Paying Not to Go to the Gymulrike/Papers/gym.pdf · 2006-09-12 · Paying Not to Go to the Gym By S...

Paying Not to Go to the Gym

By STEFANO DELLAVIGNA AND ULRIKE MALMENDIER

How do consumers choose from a menu of contracts We analyze a novel datasetfrom three US health clubs with information on both the contractual choice and theday-to-day attendance decisions of 7752 members over three years The observedconsumer behavior is difficult to reconcile with standard preferences and beliefsFirst members who choose a contract with a flat monthly fee of over $70 attend onaverage 43 times per month They pay a price per expected visit of more than $17even though they could pay $10 per visit using a 10-visit pass On average theseusers forgo savings of $600 during their membership Second consumers whochoose a monthly contract are 17 percent more likely to stay enrolled beyond oneyear than users committing for a year This is surprising because monthly memberspay higher fees for the option to cancel each month We also document cancellationdelays and attendance expectations among other findings Leading explanations forour findings are overconfidence about future self-control or about future efficiencyOverconfident agents overestimate attendance as well as the cancellation proba-bility of automatically renewed contracts Our results suggest that making infer-ences from observed contract choice under the rational expectation hypothesis canlead to biases in the estimation of consumer preferences (JEL D00 D12 D91)

ldquoSaturday 31 December New Yearrsquos Res-olutions I WILL [] go to the gym threetimes a week not merely to buy sandwichrdquo

Bridget Jonesrsquos Diary A Novel

ldquoMonday 28 April [] Gym visits 0 noof gym visits so far this year 1 cost ofgym membership per year pound370 cost ofsingle gym visit pound123 (v bad economy)rdquo

Bridget Jones The Edge of Reason

Many firms offer consumers a menu of con-tracts Cellular phone users choose combina-

tions of monthly airtime minutes and pricesCredit card users choose between teaser rateoffers and contracts with a constant interest rateA large literature in industrial organization an-alyzes the profit-maximizing contract design(Jean Tirole 1988) A standard assumption inthis literature is that consumers have rationalexpectations about their future consumption fre-quency and choose the utility-maximizingcontract

In this paper we provide evidence that thismay not always be the case We present a noveldataset from three US health clubs that allowsus to analyze the contractual choices of consum-ers in light of their actual consumption behav- DellaVigna Department of Economics University of

California at Berkeley 549 Evans Hall Berkeley CA94720 and National Bureau of Economic Research (e-mailsdellavieconberkeleyedu) Malmendier Graduate Schoolof Business Stanford University 518 Memorial Way Stan-ford CA 94305 and NBER (e-mail malmendier_ulrikegsbstanfordedu) Earlier versions of this paper weredistributed under the titles ldquoSelf-Control in the Market Evi-dence from the Health Club Industryrdquo and ldquoOverestimatingSelf-Control Evidence from the Health Club Industryrdquo Wethank Doug Bernheim (the editor) two anonymous refereesMarios Angeletos George Baker Rachel Croson Rajeev De-hejia Richard Gilbert Oliver Hart Caroline Hoxby DanielePaserman Antonio Rangel Emmanuel Saez Andrei ShleiferJeremy Tobacman Klaus Wertenbroch Justin Wolfers and inparticular Edward Glaeser Lawrence Katz and David Laib-son for their comments We also thank the participants of

seminars in CREST Dartmouth College Harvard UniversityUniversity of Michigan Ann Arbor Northwestern UniversityNew York University UC Berkeley UCLA the Ente Einaudithe SITE 2001 and 2002 and at the Eastern Economic Asso-ciation 2002 for their comments We thank Bryan S Grahamfor providing Mathematica code and Tobias Adrian AugustinLandier and Sendhil Mullainathan for inspiring conversationsat the beginning of this project Tatyana Deryugina TriciaGlynn Burak Guner Camelia Kuhnen Scott McLinn BorisNenchev Nikita Piankov and Justin Sydnor provided excel-lent research assistance For financial support DellaVignathanks the Bank of Italy and Malmendier thanks the GermanAcademic Exchange Service (DAAD)

694

ior The dataset contains information both onthe type of membership and the day-to-day at-tendance decisions of 7752 health club mem-bers over three years We find that consumerschoose a contract that appears suboptimal giventheir attendance frequency In addition low-attendance consumers delay cancelling this con-tract despite small transaction costs

Our empirical analysis exploits the presenceof a contractual menu Consumers can choosebetween two flat-rate contractsmdasha monthlycontract and an annual contractmdashand a pay-per-visit option The monthly contract is automati-cally renewed from month to month until theconsumer cancels The annual contract insteadexpires after 12 months unless the consumerexplicitly renews it The variation in the per-usage pricing and in the renewal proceduresallows us to identify several puzzling features ofconsumer behavior

First consumers who choose a monthlymembership of over $70 per month pay onaverage 70 percent more than they would underthe pay-as-you-go contract for the same number

of visits Eighty percent of the monthly mem-bers would have been better off had they paidper visit for the same number of visits

Second consumers who choose the monthlycontract are 17 percent more likely to stay en-rolled beyond one year than users choosing theannual contract This is surprising becausemonthly members pay higher fees for the optionto cancel each month This result occurs eventhough high-attendance users sort into the an-nual contract at enrollment

These and additional empirical findings(summarized in Table 1) are hard to reconcilewith standard preferences and beliefs We ex-plore potential explanations including hightransaction costs of payment per usage riskaversion underestimation of costs of attendanceand of cancellation time inconsistency naiveteabout the time inconsistency and persuasion byhealth club employees

In our view the most parsimonious explana-tions are those allowing for overconfidence (na-ivete) Consumers overestimate for exampletheir future self-control or their future efficiency

TABLE 1mdashEMPIRICAL FEATURES AND POSSIBLE EXPLANATIONS

Standardmodel

Trans costsof paymentper usage

Membershipbenefits per

usageLimitedmemory

Timeinconsist

withsophistication

Time inconsistwith naivete

Overestimationof futureefficiency Persuasion

Finding 1Price per average attendance

$1727Distaste of

pay perusage

Membershipbenefits

Commitment Commitmentoverestimationof attendance

Overestimationof attendance

Pressure ofsalesman

Finding 2Average attendance in months

2ndash4 higher in annual thanmonthly contract

Sorting atenrollment








Finding 3Users predict 950 monthly

visits actual monthly visitsare 417



Finding 4Interval between last

attendance and termination231 full months

Distaste ofpay perusage

Membershipbenefits

Forget tocancel

Overestimationof cancellation


Pressure ofsalesman

Finding 5Survival probability after 14

months 17 percent higherfor monthly than for annualcontract

Forget tocancel



Pressure ofsalesman

Finding 6Average attendance 27 percent

higher in second year forannual contract

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Finding 7Decreasing average attendance

over time in monthlycontract

Forget tocancel



Pressure ofsalesman

Finding 8Positive correlation of price

per average attendance andinterval between lastattendance and termination

Heterogeneity innaivete

Heterogeneity inoverconf

695VOL 96 NO 3 DELLAVIGNA AND MALMENDIER PAYING NOT TO GO TO THE GYM

in pursuing costly activities This leads to over-estimation of attendance and of cancellation inautomatically renewed contracts As an alterna-tive explanation persuasion by health club em-ployees can explain most findings

In a simple yet economically significant de-cisionmdashenrollment and attendance in a healthclubmdashconsumers deviate systematically fromthe optimal contractual choice In the healthclubs of our sample the average nonsubsidizeduser chooses the monthly contract and by doingso forgoes savings of about $600 per member-ship out of a total amount of about $1400 paidto the health club The results of this study arelikely to generalize to the 328 million Ameri-cans who exercise in one of the 16983 UShealth clubs Therefore both in terms of monetarymagnitude and in terms of population involvedthe nonstandard behavior has a significant eco-nomic impact Our findings are also consistentwith findings on consumer behavior in the creditcard industry (Haiyan Shui and Lawrence MAusubel 2004) and employee choice of 401(k)plans (Brigitte C Madrian and Dennis F Shea2001)

The analysis of consumer behavior is just thefirst step toward a better understanding of in-dustries where consumers display nonstandardpreferences or beliefs Profit-maximizing firmsshould respond to the nonstandard features ofconsumer behavior in their contract design Thisis the central theme of the growing literature onbehavioral industrial organization (DellaVignaand Malmendier 2004 Kfir Eliaz and RanSpiegler forthcoming Xavier Gabaix andDavid Laibson forthcoming Paul Heidhuesand Botond Koszegi 2005) surveyed in GlennEllison (forthcoming) The large effect of smallcancellation costs on renewal rates may explainthe high frequency of contracts with automaticrenewal in the newspaper credit card and mail-order industry The findings have implicationsalso for the design of flat-rate pricing (EugenioJ Miravete 2003) In DellaVigna and Mal-mendier (2004) we explore the implications forfirm pricing of a leading explanation of ourresults overconfidence about self-control

Our findings suggest caution in making infer-ences about consumer preferences from ob-served choices of products (Igal Hendel andAviv Nevo 2004) or contracts (Miravete andLars-Hendrik Roller 2003) when actual con-sumption is unobserved Inferences made under

the assumption of rational expectations can leadto significant bias For example we would haveconcluded that monthly members attend on av-erage at least twice a week This erroneousconclusion would have overstated the impact ofhealth club enrollment on health outcomes

Finally our findings have implications for thepolicy debate on obesity (David M Cutler et al2003) Subsidizing enrollment in health clubs islikely to have only small effects on obesityrates given the low average attendance ofmembers

The remainder of the paper is organized asfollows In Section I we introduce the mainfeatures of the health club dataset In Section IIwe develop predictions about the contractualchoice at enrollment and test the predictionsempirically In Section III we present a similaranalysis of the contractual choice and consump-tion behavior over time Section IV discussespossible explanations for the empirical findingsSection V concludes

I Health Club Dataset

A Health Club Industry

As of January 2001 16983 clubs were oper-ating in the United States The industry reve-nues for the year 2000 totalled $116 billionThe memberships in the same period was 328million up from 174 million in 1987 Fifty-onepercent of the users were members in commer-cial health clubs while 34 percent were mem-bers in nonprofit facilities Only the marketleader Bally Total Fitness with $1007 millionin revenues and 4 million members is publiclytraded Few companies operate in more than tenstates Ownership concentration is in the tenthpercentile of US industries

B Dataset

We collected a new panel dataset from threehealth clubs located in New England which welabel clubs 1 2 and 3 The dataset containsinformation on the contractual choices and theday-to-day attendance of users who enrolledafter April 1 1997 The sample period is April1997 through July 2000 for club 1 and April1997 through February 2001 for clubs 2 and 3The day-to-day record of usage is made avail-able by the technology regulating the access to

696 THE AMERICAN ECONOMIC REVIEW JUNE 2006

these health clubs described below The panelof contractual choices comes from the billingrecords Each entry in the accounting data spec-ifies the price paid for the transaction and afour-letter code This code allows us to track themembership typemdashstandard student familycorporatemdashas well as details like the subsidiz-ing company (if any)

Several companies located near the clubssubsidize their employeesrsquo attendance Forthese corporate members the health club re-ceives part of the membership payments di-rectly from the firms with the remainder beingpaid by the members The health club informsthe companies periodically about the number ofemployees enrolled and their attendance Thiscreates incentives for the health club to recordattendances accurately or possibly to over-record them

C Contractual Menu

We conducted a survey of the 97 health clubsin the Boston metropolitan area to document thecontract design in the industry1 Health clubsoffer up to three options 87 clubs offer amonthly contract and a monthly fee is automat-ically debited each month to a credit card orbank account until the user cancels 90 clubsoffer an annual contract Both monthly and an-nual contracts have an initiation fee but no feeper visit Finally 82 clubs offer a pay-per-visitoption often in the form of a ten-visit passHealth clubs 1 and 2 in our sample offer thethree types of contract with the following addi-tional features2

The monthly contract has a monthly fee rang-ing between $70 (discounted level) and $85(standard level) Noncorporate users also payan initiation fee ranging from $0 (in promo-tional periods) to $150 Corporate users gen-erally pay an out-of-pocket monthly feebetween $19 and $65 depending on the sub-

sidy paid by their company and no initiationfee Cancellation can be done in person at theclub or by sending a written note3 If cancel-lation takes place before the 10th of themonth no further fees are due and the userscan attend until the end of the month Mem-bers who cancel after the 10th have to pay thefee for the next month and can attend until theend of the following month

The annual contract charges up front tentimes the applicable monthly fee eg $850for a standard membership4 Users thus get adiscount of 2 months out of 12 in exchangefor a yearly commitment The initiation fee isthe same as under the corresponding monthlycontract At the end of the year the contractexpires and members who wish to stay en-rolled have to sign up again either for anannual or for a monthly contract In order toencourage renewal the club sends out a re-minder card one month before the contractexpires

The pay-per-visit system offers two optionseither to pay $12 per visit or to purchase aten-visit pass for $100 Transaction costs forthe ten-visit pass are small Users providebasic demographic information and receive acard for ten visits Unfortunately attendanceis not tracked for the pay-per-visit users

Users of club 3 face the same menu of con-tracts with lower prices and slightly differentservices The monthly fee ranges from $13 to$52 and the initiation fee is at most $50 Theannual fee in the annual contract equals tentimes the corresponding monthly fee The pay-per-visit options are a $10 fee per visit and a$80 pass for ten visits

Under all types of membership users receivecards they have to deposit in a basket at thefront desk when they enter While they areexercising a health club employee swipes them(marks the visit for the ten-visit passes) andusers pick them up when they exit This methodguarantees a high recording precision even dur-ing peak hours The three contracts give right tothe same services ie a temporary locker tow-1 For details on the survey see DellaVigna and Mal-

mendier (2004)2 Contracts for one to six months with automatic expi-

ration are also available We do not include them in ouranalysis since they are typically targeted toward occasionalsummer users We also remove from the sample freelimited-time memberships that are occasionally given toemployees of the subsidizing companies

3 Some users cancel by discontinuing the payments tothe health club

4 The annual fee can be paid in three installments due inthe first six months


els5 and access to the equipment Also both themonthly and the annual contract allow membersto ldquofreezerdquo (suspend) their membership forthree months per year6 Users with a monthlycontract do not have to pay their monthly feeduring the freezing period Annual members getadditional usage time after the original 12months

D Sample Construction

We match the information on attendance andon contract choice in the three clubs to form alongitudinal dataset with monthly observationscovering the period from April 1997 to July2000 (club 1) and to February 2001 (clubs 2 and3) Our analysis focuses on enrollment spells Aspell starts whenever an individual enrolls (orreenrolls) in a club and ends whenever the in-dividual quits We define spells to be censoredif either the enrollment is ongoing at the end ofthe sample period or the individual switches toa short-term contract or receives a promotionalmembership Accordingly spells are completedif the individual cancels the membership (undera monthly contract) or if the membership ex-pires (under an annual contract) within the sam-ple period Individuals have multiple spells ifthey quit the club and reenroll at some laterdate

The initial sample includes 10175 individu-als We drop individuals who were never en-rolled in either a monthly or an annual contract(1867 individuals) We eliminate individualswith data inconsistencies (49 individuals) Wealso exclude users with a family membership toavoid issues regarding the joint consumption ofthe services (247 individuals) Finally in orderto limit the sample to first-time users of theseclubs we drop users who had a free or a sea-sonal membership before they chose a monthlyor an annual contract (260 individuals) (Addi-tional information on the dataset construction isavailable in the Data Appendix)

This leaves us with a sample of 7752 indi-viduals and 8273 enrollment spells In the pa-per we consider only the first enrollment spellfor each individual As row 1 of Table 2 shows

club 1 has 22 percent more members than club2 and more than twice as many members asclub 3 The percentage of completed spells issimilar across the clubs above 60 percent Ofthe 7752 individuals enrolled in any club 89percent choose a monthly membership as theirfirst contract Health club members rarelychange the type of contract they initially enrollin In addition to the whole sample we also usethe sample ldquono subsidyrdquo which includes onlyunsubsidized memberships We consider amembership to be unsubsidized if over thewhole spell the average out-of-pocket fee ex-ceeds $70 per month for enrollment in amonthly membership and $700 per year ($58per month) for enrollment in an annual mem-bership This smaller sample includes 1070 in-dividuals (14 percent of the full sample)

E Descriptive Statistics

In clubs 1 and 2 (columns 1 and 2) theaverage amount spent per spell is about $550and the average fee per month ranges between$44 and $52 For corporate users these areout-of-pocket payments and do not include sub-sidies paid by the sponsoring firms Theamounts are substantially lower in club 3 (col-umn 3) since the contracts are cheaper andsubstantially higher in the sample ldquono subsidyrdquo(columns 7 and 8) Across all clubs (column 4)the initiation fee averages $4 and is paid by only14 percent of users Individuals with a monthlycontract attend on average four times permonth and individuals with an annual contractattend on average 44 times per month Atten-dance in club 1 (column 1) is somewhat higherthan in the other clubs Freezing of a contract israre in all the clubs The bottom part of Table2 displays the available demographic controlsUsers are somewhat more likely to be male thanfemale and are on average in their early thirtiesCorporate memberships account for 50 percentof the sample while student memberships ac-count for only 2 percent

II Contract Choice at Enrollment

A Predictions of the Standard Model

We set up a model of contract choice andhealth club attendance We assume that health

5 Towels are not included in memberships in club 36 Monthly users can also quit for up to three additional

months without repaying the initiation fee


club attendance involves immediate effort costsand delayed health benefits and that the effortcosts are uncertain ex ante In particular costs

can be high (c c) or low (c c) and indi-viduals differ in the ex ante probability thatcosts will be high A contract (L p T) gives

TABLE 2mdashDESCRIPTIVE STATISTICS

Sample AllSample All Sample No subsidy

Club 1 Club 2 Club 3 All clubsAll clubs All clubs

Allcontr

(1)

Allcontr

(2)

Allcontr

(3)

Allcontr

(4)

Firstcontractmonthly

(5)

Firstcontractannual

(6)


(7)

Firstcontractannual

(8)

Number of spellsTotal 3495 2866 1391 7752 6875 877 866 204Completed spells 2431 1825 990 5246 5246 509 581 112

Total amount in $ 55830 55150 31408 51196 49840 61825 91802 102256(50052) (55150) (30418) (50052) (50494) (45071) (69958) (53689)

N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204Initiation fee 635 191 289 409 388 574 1468 1765

(2664) (1191) (1303) (2023) (1951) (2510) (4188) (4557)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Average fee per monthMonthly contract 5214 4904 3127 4222 4712 5598 7856 7360

(1857) (1909) (1097) (1922) (1919) (2058) (503) (1578)N 3185 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20

Annual contract 4819 4433 2413 4301 4699 4257 7012 6627(1564) (1708) (875) (1745) (1510) (1764) (454) (403)

N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204Average attendance per month

Monthly contract 413 398 376 401 400 449 393 520(392) (376) (369) (382) (382) (377) (376) (429)

N 3138 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20Annual contract 457 422 420 437 571 422 726 435

(398) (408) (395) (401) (427) (396) (350) (395)N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204

Contract choice per spellMonths with monthly contract 903 695 894 898 1008 042 1167 050

(827) (903) (884) (866) (857) (208) (887) (226)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Months with annual contract 155 197 142 168 015 1368 007 1492(467) (578) (483) (514) (150) (732) (105) (786)

N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204Freezing 026 031 018 026 029 005 035 004

(094) (114) (072) (099) (104) (038) (120) (032)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Female 044 048 047 046 048 034 038 035(050) (050) (050) (050) (050) (047) (049) (048)

N 3487 N 2866 N 1391 N 7744 N 6875 N 876 N 866 N 204Age at sign-up 3071 3151 3508 3179 3150 3406 3312 3442

(844) (891) (930) (891) (878) (963) (975) (1086)N 3293 N 2745 N 1316 N 7354 N 6523 N 831 N 812 N 193

Corporate member 043 061 043 050 050 052 017 016(050) (049) (050) (050) (050) (050) (037) (037)

N 3495 N 2866 N 1391 N 7752 N 7079 N 877 N 866 N 204Student 005 000 000 002 002 001 000 000

(021) (005) (005) (015) (015) (012) (005) (007)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Notes Standard deviation in parentheses An enrollment spell starts whenever an individual enrolls in the club and endswhenever the individual quits or is censored The sample ldquono subsidyrdquo consists of the spells in which the average adjustedmonthly fee is at least $70 if the spell starts with a monthly contract and at least $58 if the spell starts with an annual contractThe spells in column ldquofirst contract monthlyrdquo start with a monthly contract The spells in column ldquofirst contract annualrdquo startwith an annual contract ldquoAverage price per monthrdquo refers to the out-of-pocket fee in the case of corporate users


customers the right to exercise for a fee p andfor T periods (days) once the flat fee L is paidWe assume that consumers can choose betweenflat-fee contracts (like the monthly and annualcontract) with p 0 and pay-per-visit contractswith L 0 We summarize here the results oncontract choice for the case of standard prefer-ences and rational beliefs The derivation is inthe working-paper version (DellaVigna andMalmendier 2002)

Flat Rate versus Pay per UsagemdashWe con-sider first the choice at enrollment between aflat-rate contract (L 0 T) and a pay-per-visitcontract (0 p T) Denote by the daily discountfactor and by EF[v] the expected number ofvisits (over T days) under the flat-rate contract

Prediction 1 (price per expected attendanceat enrollment) For agents who choose a flat-rate contract

(1)L

EF vaT p

The factor a(T) (1 )T(1 T) is atime-adjustment coefficient due to the fact thatthe flat fee L is paid up front and the per-visitfee p is paid every period between 1 and T Forsmall T such as T 30 under the monthlycontract a(T) is approximately 1 Equation (1)says that payment per expected visit under theflat-rate contract should be smaller than theper-visit-fee p Intuitively only consumers whoattend frequently should choose the flat-ratecontract Suppose instead that a consumer un-der the flat-rate contract attends infrequentlyenough that the price per expected visit LEF[v]is greater than the per-visit-fee p If this con-sumer switched to the pay-per-visit contractwithout changing state-contingent attendanceshe would have higher utility Reoptimizing theattendance choices she must be even better off

Annual versus Monthly ContractmdashThe an-nual contract A requires a yearly commitmentThe monthly contract M offers the option tocancel in any period but charges a higher fee permonth Consumers who anticipate a highenough probability of being high-cost types(c c) prefer the monthly contract for its flex-ibility Users who believe that they will below-cost types prefer the annual contract The

users who select the annual contract thereforeare more likely to be frequent users In Predic-tion 2 we use attendance in the initial monthsE[v] (before the selective exit) as a measure ofthe likelihood to be a frequent user

Prediction 2 (attendance of monthly and an-nual members) The average initial attendanceof annual members is higher than the averageinitial attendance of monthly members

EA v EM v

A third test for the standard model is whetherconsumers have rational expectations abouttheir attendance

Prediction 3 (forecast of attendance) The av-erage forecast of attendance equals the averageactual attendance

B Empirical Analysis

We test Prediction 1 using the sample ofusers enrolled in an unsubsidized flat-rate mem-bership in clubs 1 and 2 We analyze separatelyusers in club 3 given the lower fee per visitAs the benchmark measure of price per visitwe use the price per visit under the ten-visitpass $10 rather than the $12 visit-by-visitfee the ten-visit pass is cheaper for users witha monthly or annual contract given their at-tendance frequency7

Monthly ContractmdashFor users initially en-rolled in a monthly contract we compute theprice per expected attendance for each monthWe limit the analysis to the first six months oftenure to target inexperienced users We use thesample ldquono subsidyrdquo (866 individuals) to ensurecomparability to standard health clubs with nocorporate subsidy

The first column in Table 3 reports the aver-age monthly fees in months one through six

7 The (hypothetical) average price per average atten-dance from using the ten-visit pass given the distributionof attendance for users enrolled with the monthly and theannual contract is $1091 The benefits of a lower pricerelative to the $12-per-visit fee outweigh the losses fromunused coupons for these users The single-visit fee of$12 is targeted toward one-time users such as travellingbusinessmen


with standards errors in parentheses The sam-ple for month t consists of users who initiallyenrolled in a monthly contract and have had acontinuous history of membership up to montht under either a monthly or an annual contractConsumers drop out of the sample when theycancel or are censored For users who switch toan annual contract the monthly fee is the monthlyshare of the annual fee The average monthly fee

exceeds $80 in all months except in the joiningmonth which is typically prorated and in month 3a promotional free month for 186 percent of thesample The average number of visits for users inthe tth month of tenure (column 2) declines from546 in month 2 to 432 in month 6 (Month 1covers only part of a month)

The third column in Table 3 presents the ratioof the average fee in month t (column 1) and the

TABLE 3mdashPRICE PER AVERAGE ATTENDANCE AT ENROLLMENT

Sample No subsidy all clubs

Average priceper month

(1)

Average attendanceper month

(2)

Average priceper average attendance

(3)

Users initially enrolled with a monthly contract

Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866

Users initially enrolled with an annual contract who joined at least14 months before the end of sample period

Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145

Notes Standard errors in parentheses Standard errors for ldquoaverage price per average atten-dancerdquo measure computed using the bivariate Delta method The number of observations isdenoted by N An enrollment spell starts whenever an individual enrolls in the club and endswhenever the individual quits or is censored The sample ldquono subsidyrdquo consists of the spellsin which the average adjusted monthly fee is at least $70 if the spell starts with a monthlycontract and at least $58 if the spell starts with an annual contract The sample for the t-thmonth includes spells that are ongoing not frozen and not miscoded at month t For thesix-month period the sample includes spells that are ongoing not frozen and not miscodedin at least one month in the period For the one-year period in the annual contract the sampleincludes only spells that started at least 14 months before the end of the sample period andthat were not prematurely terminated because of medical reasons or relocation The ldquoaveragepricerdquo in period t is the average fee across people enrolled in period t The ldquoaverageattendancerdquo in period t is the average number of visits across people enrolled in period t Themeasure in column 3 is the ratio of the measure in column 1 and the measure in column 2


average attendance in month t (column 2) Thisratio is the estimated price per expected atten-dance for month t (LEF[v])a(T) in Prediction 1In each of the six months we reject the hypoth-esis that the price per expected attendance issmaller than $10 (or than $12) The estimateranges between $14 and $16 in the first threemonths and is higher than $17 in the subsequentthree months As a summary measure we com-pute the ratio of average monthly payment (col-umn 1) and average monthly attendance(column 2) in the first six months across allindividuals8 The resulting price per averageattendance in the first six months of enrollmentequals $1727 well above $10 (or $12)

In addition to averages we consider also thedistribution of these measures in the first sixmonths (Table 4) We measure the price perattendance as the ratio of total attendance over

total payment in the first six months of mem-bership in a monthly contract (column 2) Only20 percent of the individuals pays less than $10per visit The remaining 80 percent would havesaved money choosing the pay-per-visit con-tract holding constant the number of visits

Annual ContractmdashWe also test Prediction 1on the users who chose an annual contract atenrollment We use the sample ldquono subsidyrdquofurther restricted to users who joined the club atleast 14 months before the end of the sampleperiod (145 individuals) This ensures that weobserve the annual contract in its entirety9

The bottom row of Table 3 presents the esti-mation results The average monthly share ofthe annual fee for the first year (column 1)adjusted for discounting is $663210 The aver-age number of monthly visits in the first year

8 For each individual we compute the average over allavailable months until the sixth with the exception ofmiscoded months and months with freezing When averag-ing across individuals we weigh all individuals equallyindependent of tenure

9 We exclude three annual contracts that are terminatedbefore the twelfth month Health clubs are required toaccept cancellations for medical reasons or for relocationmore than 25 miles from the clubs

10 We use a daily discount factor of 09998 implying anadjustment factor T(1 )(1 T) equal to 1037

TABLE 4mdashDISTRIBUTION OF ATTENDANCE AND PRICE PER ATTENDANCE AT ENROLLMENT


First contract monthlymonths 1ndash6

(monthly fee $70)

First contract annualyear 1

(annual fee $700)

Averageattendanceper month

(1)

Price perattendance

(2)


(3)

Price perattendance

(4)

Distribution of measures10th percentile 024 773 020 59820th percentile 080 1018 080 88125th percentile 119 1148 108 1127Median 350 2189 346 196375th percentile 650 6375 608 630690th percentile 972 12173 1086 1138595th percentile 1178 20110 1316 29451

N 866 N 866 N 145 N 145

Notes The number of observations is denoted by N An enrollment spell starts whenever anindividual enrolls in the club and ends whenever the individual quits or is censored Thesample ldquono subsidyrdquo consists of the spells in which the average adjusted monthly fee is at least$70 if the spell starts with a monthly contract and at least $58 if the spell starts with an annualcontract The spells in column ldquofirst contract monthly months 1ndash6rdquo start with a monthlycontract The spells in column ldquofirst contract annual year 1rdquo start with an annual contract Thevariable ldquoprice per attendancerdquo is defined as the ratio of the average price over the averageattendance over the first period (six months for the monthly contract one year for the annualcontract)


(column 2) is 436 The resulting price per av-erage attendance (column 3) of $1522 is sub-stantially higher than $10 (or than $12) Theestimate is somewhat lower than for themonthly contract consistent with selection ofusers with higher expected attendance into theannual contract (Prediction 2) Table 4 showsthe distribution across users of attendance (col-umn 3) and of the price per attendance (column4) in the first year of an annual membershipOnly 24 percent pay less than $10 per visit

Finding 1 (price per expected attendance atenrollment) Users who choose an unsubsi-dized flat-rate contract pay a price per averageattendance of over $17 in the monthly contractand over $15 in the annual contract The shareof users who pay ex post less than $10 per visitis 20 percent in the monthly contract and 24percent in the annual contract

Size of the EffectmdashAs a monetary measureof the deviation from the standard model formonthly and annual memberships we computethe difference between actual expenses over thewhole enrollment spell and imputed expensesfor the same number of attendances with ten-visit passes11 This measure understates the sav-ings from paying per visit since the agents couldreoptimize their attendance The ldquoaverage lossrdquomeasure is positive if the user would have savedmoney purchasing ten-visit passes and negativeotherwise We use the sample ldquono subsidyrdquo forspells that start before October 1997

The average loss per spell is $614 for agentsinitially enrolled in a monthly contract Thisamount is 43 percent of the overall $1423 spenton the health club membership For agents ini-tially enrolled in an annual contract there is asmall insignificant gain of $1

The observed deviation from the standardmodel has large monetary consequences for us-ers in the monthly contract For users in theannual contract the automatic expiration mod-erates the possible losses

RobustnessmdashWe now check the robustnessof Finding 1

1 Sample Thus far we have restricted at-

tention to the unsubsidized sample and pooledthe results across clubs We now include allusers who initially chose a monthly contract anddisaggregate the results by club Separately foreach club we regress health club attendance onthe monthly fee using an Epanechnikov kernelThe measure of attendance is the average atten-dance per month in the first six months Wecross-validate club by club with a grid search tocompute the optimal bandwidth for the price12

In club 1 (Figure 1A) the average monthlyattendance from the kernel regression lies be-tween three and five and is increasing in pricealthough the estimates are not very smoothgiven the small bandwidth suggested by thecross-validation We use the average attendancefrom the kernel regression to compute the ratioof price and average attendance for each level ofprice (Figure 1B) The price per average atten-dance is significantly higher than $10 for userspaying a monthly fee in excess of $53 Theestimates for club 2 are comparable (Figures 1Cand 1D) and somewhat smoother given thelarger optimal bandwidth In club 3 the price peraverage attendance is higher than the per-visitfee of $8 for users paying a fee in excess of $46(Figure 1F)

2 Underrecording of attendance The highprice per attendance could result from underre-cording of attendance due to a faulty computersystem or moral hazard problems with the staffHealth club employees may also seek to avoidqueues of users waiting to swipe The threehealth clubs in our sample had incentives toaddress these problems since they provide re-ports of attendance to the corporations subsidiz-ing employee memberships They therefore putin place one of the most advanced and reliablesystems to track attendance in the industry Un-like in most clubs a front-desk employee col-lects the cards from the members and swipesthem while the member is exercising There-fore card swiping does not generate queues Wealso witnessed the procedure if a member hasforgotten the card the employee looks the nameup in the computer and records the attendanceThus while errors may occur in both direc-tionsmdashfailure to swipe and double swipingmdashthe health club data used in our analysis areunusually accurate

11 This measure takes into account the potential lossassociated with not using fully a ten-visit pass 12 Adrian Pagan and Aman Ullah (1999) pp 110ndash20


FIGURE 1 AVERAGE ATTENDANCE AND PRICE PER AVERAGE ATTENDANCE (KERNEL REGRESSIONS)

Notes Point estimates and 95-percent confidence intervals plotted The sample is all individuals initially enrolled with amonthly contract The individual price variable is the average price over the first six months The individual attendancevariable is the average attendance over the first six months Figures 1A 1C and 1E show a kernel regression of attendanceon price using an Epanechnikov kernel The bandwidth is determined by cross-validation with a grid search separately foreach club Figures 1B 1D and 1F show the ratio of the price and the expected attendance predicted for that price using thekernel regression Confidence intervals are derived using the Delta method


As supporting evidence we can test whetherrandom events such as computer crashes oridiosyncratic laziness of employees affect sub-stantially the accuracy of the attendance recordWe calculate the fraction of members attendingon each day in the sample and regress it on a setof controls 6 day-of-the-week dummies 11month dummies 3 year dummies and 15 holi-day dummies If recording precision is highlyvariable the R2 of this regression should be lowThe R2 of the regression for club 1 instead is ashigh as 08785 with the day-of-the-week dum-mies explaining most of the variance The re-gression for clubs 2 and 3 yield an even higherR2 of 0891513 The high explanatory power ofthese regressions suggests that daily variation inrecording precision is limited

3 Ex post subsidies Some HMOs reimbursemembers partially for health club expenses Tothe extent that these reimbursements make theannual and the monthly contract cheaper rela-tive to the pay-per-visit contract they induceusers to choose flat-rate contracts However theHMOs in the state where the three clubs operateoffer discounts either on the initiation fee only orto both flat-rate and pay-per-usage contracts14

4 Membership benefits Consumersrsquo choiceof the monthly or annual contract could be dueto contract-specific membership benefits Theonly benefit not available under the per-visitpayment though is the option to rent an over-night locker at an extra fee and only 94 percentof the users ever rent a locker If we excludethese users the results on price per averageattendance for the monthly contract do not vary

Overall we observe a robust deviation fromPrediction 1 Nonsubsidized users enrolled incontracts with flat fees pay a price per averageattendance that is significantly higher than theper-visit price available as an alternative con-tract The result is robust to the type of contract(monthly or annual) the sample (the amount ofsubsidy) and the club considered The resultsdo not appear to depend on measurement errorex post subsidies or unobserved benefits Thedeviation from Prediction 1 is large unsubsi-

dized members of a monthly contract pay 70percent in excess of the $10 fee

To test Prediction 2 on the initial sortingbetween the monthly and the annual contractswe compare the average number of visits inmonths 2 3 and 4 of tenure for individualsinitially enrolled in the monthly and in the an-nual contract15 Given that the price per visit pis zero for both contracts differences in atten-dance should reflect differences in the expectedfuture attendance cost Column 1 of Table 5 re-ports the results for the whole sample In eachmonth expected attendance is higher under theannual than under the monthly contract andsignificantly so in months 3 and 4 Overallaverage attendance in months 2 to 4 is 10 per-cent higher under the annual contract The mag-nitude of this difference is comparable tovariation in average attendance by age groupsand by gender When we break down the sam-ple into 24 age-gender-month cells average at-tendance is higher under the annual contract in

13 Detailed results are available in DellaVigna and Mal-mendier (2002 Appendix Table 1)

14 We report the results in Appendix Table 3 in Della-Vigna and Malmendier (2002) We thank Nancy Beaulieufor providing the list of HMOs

15 We exclude the first month because attendance isprorated over the number of effective days of membershipand the prorating procedure is slightly different for theannual and the monthly contract We do not extend thecomparison to months after the fourth since users whoexperience a high cost can quit under the monthly contractbut not under the annual contract

TABLE 5mdashAVERAGE ATTENDANCE IN MONTHLY AND

ANNUAL CONTRACTS

(Sorting)

Average attendance during the n-thmonth since enrollment

Sample All clubs

Month 2 Month 3 Month 4

Monthly contract 5507 5005 4614(00668) (00696) (00709)

N 6219 N 5693 N 5225Annual contract 5805 5629 5193

(01885) (01934) (01913)N 862 N 841 N 817

Notes Standard errors in parentheses The number of ob-servations is denoted by N An enrollment spell starts when-ever an individual enrolls (or reenrolls) in the club and endswhenever the individual quits or is censored The spells inrow ldquomonthly contractrdquo start with a monthly contract Thespells in row ldquoannual contractrdquo start with an annual contractThe sample in month n includes spells that are ongoing notfrozen and not miscoded


20 cells out of 24 Even after controlling forsome heterogeneity individuals with higher at-tendance are more likely to choose the annualcontract at enrollment

Finding 2 (attendance of monthly and annualmembers) Average attendance in months 2ndash4is 10 percent higher under the annual contractthan under the monthly contract

While consumersrsquo choice between flat-ratecontracts and a per-visit fee is hard to explain inthe standard framework (Finding 1) theirchoice between the monthly and annual contract(Finding 2) is consistent with standard prefer-ences and beliefs Consumers sort according tothe expected attendance

Finally we elicit the expectations of healthclub members about their future attendance us-ing a survey of 48 randomly chosen respondentsinterviewed in a mall16 The mall is not near ahealth club so the respondents are not selectedon health club attendance We ask the ones thatreport to be members or to attend a health clubhow often they expect to visit their health clubin the next month September17 This questionattempts to measure directly whether healthclub users have rational expectations Althoughwe do not observe actual attendance amongthese 48 survey respondents it is unlikely todiffer substantially from attendance in our data-set which is very robust across demographicsubgroups Across 24 (gender)(club)(age)subgroups the average monthly attendance overthe membership is lower than 475 visits for 23out of 24 groups with an overall average of417 monthly visits

Finding 3 (forecasts of attendance) The av-erage forecasted number of monthly visits 950(se 066) is more than twice as large as av-erage attendance 417

The overestimation displayed by the subjectsmatches Finding 1 If health club consumersexpect to attend 95 times per month theyshould indeed choose a flat-rate contract ratherthan paying per visit

We also present the subjects with the follow-ing scenario Suppose that based on your pre-

vious experience you expect to attend onaverage five times per month (about once aweek) if you enroll in a monthly membershipYou plan to attend the health club throughoutthe next year Would you choose a monthlycontract with a monthly fee of $70 per month orten-visit passes for $100 (each visit costs $10)This question attempts to measure whether us-ers endowed with realistic expectations aboutattendance would still overwhelmingly chooseflat-rate contracts In the hypothetical scenario18 consumers out of 48 prefer the monthlycontract and 30 prefer the ten-visit pass Withrealistic expectations about attendance there-fore the majority prefers to pay per visit

These findings suggest that health club mem-bers have unrealistic expectations about theirfuture attendance One should take responses tohypothetical questions with caution howeverparticularly because the survey sample differsfrom the health club sample

III Contract Choice over Time


In the previous section we analyzed consum-ersrsquo initial choice of membership contract Inthis section we compare the renewal decisionsof monthly and annual members We take ad-vantage of two differences in the renewal pro-cedure between the two flat-rate contracts Firstthe renewal default differs The monthly con-tract is automatically renewed and requires a(small) effortmdashsending a letter or cancelling inpersonmdashin order to discontinue the member-ship The annual contract automatically expiresafter 12 months and cancellation requires noeffort Second members with a monthly con-tract can cancel at any month while memberswith an annual contract are committed for ayear We evaluate the impact of these differ-ences on cancellation lag survival probabilitiesand average attendance over time in a simplesetup with standard preferences and beliefs (de-tails are in DellaVigna and Malmendier 2002)

CalibrationmdashWe illustrate the effect of therenewal default on cancellation with the follow-ing calibration Consider two agents with iden-tical preferences and identical effort costs ofattendance One is enrolled in the monthly con-tract the other in the annual contract At the end

16 The interviews were done in August 2002 in WalnutCreek California

17 In our sample average attendance in September is 5percent lower than over the rest of the year


of the contractual period each consumer can ei-ther renew with a monthly or an annual contractor switch to the pay-per-visit contract (which isequivalent to dropping out) Denote with s the(possibly negative) daily savings from switchingto the pay-per-visit contract which we assume tobe deterministic18 The savings s are decreasing inthe future health club attendance For example amember with a monthly fee of $70 who expectsnot to attend any more has s $7030 $233Denote by the daily discount factor and by k theone-time effort cost of cancellation

Under the annual contract this cost is zeroand the agent drops out if s(1 ) 0 thatis for s 0 Under the monthly contract thecost k is stochastic with iid draws each period(day) from the cdf F In each period the agentcan switch to payment per visit at the realizedcost k or postpone switching The benefit ofpostponement is the option value of a lowerfuture realization of k while the cost is theforegone savings s The value function V solvesV E[max(k s V)] The solution ofthe agentrsquos dynamic programming problem is athreshold level k The agent switches to pay-ment per visit if the realized transaction cost issmaller than k Without solving for k wederive an upper bound on the expected numberof periods (days) until cancellation E[T] (1 F(k))F(k) under the assumption 1In Section IIIB we then compare the predictedE[T] with an empirical proxy Denote by k2 thebottom quintile of the cost distribution that isk2 F1(2) and denote by k the lower boundof the cost distribution Then E[T] must besmaller than max(4 [k2 k]s) The derivationis as follows For a cost realization of k2 theagents either switch to payment per visit or notIf they do switch for k k2 the expected delayis at most (1 F(k2))F(k2) 4 days If theydo not switch for k k2 revealed preferencesimply that the benefit of delaymdashbounded aboveby k2 kmdashmust be higher than the cost ofdelay E[T]s This yields the bound

In order to calibrate the upper bound for theexpected delay E[T] we make the conservativeassumptions k2 $10 (corresponding to thevalue of one hour of time on a calm day) andk 0 For these values an individual whoexpects not to attend the health club any more(s $7030 $233) delays on average nomore than max(4 10233) that is 43 days Anindividual who expects to attend four times amonth (s (70 40)30 $1) delays onaverage no more than ten days Under the stan-dard model therefore monthly members withlow expected attendance switch almost imme-diately to payment per visit The switching be-havior of monthly members is thus similar tothe one of annual members We summarize afirst prediction on contract choice over time

Prediction 4 (cancellation lags under themonthly contract) Low attenders under themonthly contract delay cancellation for at mosta few days

Survival ProbabilitymdashWe now compare therenewal behavior for monthly and annual con-tracts when both contracts are up for renewalie after 12 or 24 months The survival proba-bility Sjt is the probability that a consumerinitially enrolled in contract j (equal to Monthlyor Annual) is still enrolled in one of the flat-ratecontractsmdasheither monthly or annualmdashafter tmonths with t 12 24 For example SM12 isthe probability that a monthly member has notswitched to payment per visit by month 12Similarly SA12 is the probability that an annualmember renews with an annual or a monthlycontract after 12 months

Sorting at enrollment (Prediction 2) impliesthat users who selected into the annual contractare ex post more likely to be frequent usersThese users are more likely to renewmdasheitherwith a monthly or with an annual membershipThis increases SAt relative to SMt Cancellationcosts for the monthly contract instead act toincrease SMt relative to SAt The calibrationsabove however suggest that in a standard modelthe effect of cancellation costs is very small Wetherefore expect the sorting effect to dominate

Prediction 5 (survival probability) The sur-vival probability after one and after two years ishigher for agents who initially chose the annualmembership than for agents who initially chose

18 For simplicity we are neglecting the learning overtime about the savings s In a model with learning agentsmay wait to cancel for two reasons First as we capture inthe calibrations they may wait for a lower realization of kSecond they may wait for a lower realization of s Ourcalibrations show that the predictions are robust to the firstoption value argument Adding a second option value re-garding s is unlikely to change the predictions substantially


the monthly membership SAt SMt for t 1224

Attendance over TimemdashOver time monthlyand annual members learn about their atten-dance patterns and therefore about s Learninginduces selective exit of individuals with expost low attendance Define as stayers individ-uals initially enrolled in a flat-rate contract whodo not switch to a pay-per-visit contract afterthe first year Attendance of stayers in laterperiods should be higher than attendance of theinitial group since the low-attenders haveswitched to paying per visit In the standardmodel this prediction holds in similar form forboth the annual and the monthly contract19

Prediction 6 (expected attendance over timefor annual contract) Among users initiallyenrolled in an annual contract the expectedattendance in the second year among stayers ishigher than the expected attendance in the firstyear for the initial group

Prediction 7 (expected attendance over timefor monthly contract) Among users initiallyenrolled in a monthly contract the expectedattendance among stayers should increase frommonth to month


Cancellation LagsmdashTo test Prediction 4 weadopt a conservative measure of cancellationdelay E[T] for low attenders We measure thislag as the number of full months between thelast attendance and contract termination for us-ers with a monthly contract at the time of ter-mination For example if an agent attends thelast time on March 10 and cancels on April 5we count the 51 days between last attendance(March 10) and membership termination (April30) as one full month This is likely to under-state the true cancellation lag for low-attenderson two grounds (a) the measure does not in-clude months with low but positive monthlyattendance and (b) members may attend theclub one last time in order to cancel after a long

period of nonattendance We restrict the sampleto users who paid no initiation fee to ensureminimal costs of rejoining20

Finding 4 (cancellation lags under themonthly contract) On average 231 fullmonths elapse between the last attendance andcontract termination for monthly members withassociated membership payments of $187 Thislag is at least four months for 20 percent of theusers

Even though the transaction costs of cancel-lation are likely to be lower than $15 (time costof sending a cancellation letter or visiting theclub) users spend on average $187 in member-ship fees after their last attendance This lengthydelay is at odds with the calibrations in SectionIIIA which imply an average delay of at mostfive to ten days

Survival ProbabilitymdashTo test Prediction 5ideally we would compute the percentage ofmonthly members and of annual members stillenrolled one year after the initial enrollmentWe need to take into account however that (a)the first month in a contract is prorated so everyannual member is still enrolled in the thirteenth(calendar) month and (b) 115 percent of an-nual contracts last one additional month due topromotions We therefore define the survivalprobability as the share of members still en-rolled in a flat-rate contract at the fifteenth cal-endar month In order to estimate the survivalprobability we set survival si to 1 if individuali is enrolled in the fifteenth month since enroll-ment and 0 otherwise21 We use the followingempirical specification

(2) si 1 if si Mi BXi i 0

where i is normally distributed and Mi is adummy variably that equals 1 if the first con-tract for individual i is a monthly contract and0 otherwise The vector of controls X includesgender a quadratic function of age a dummy

19 The main difference is that for the annual contract thecomparison can be made only across years since the selec-tive exit is possible only every 12 months

20 We include users with an unsubsidized membership(monthly fee higher than $70 or annual fee higher than$700) who joined the club within a year since the start of thesample (April 1997)

21 The survival measure si 1 applies also to memberswho have temporarily quit the club but have reenrolled bythe fifteenth month since their initial enrollment


for corporate membership a dummy for studentmembership 11 dummies for the month and 4dummies for the year of enrollment We restrictthe sample to users who joined the club at least15 months before the end of the sample periodWe also drop users with missing values of acontrol variable as well as spells that are cen-sored before the fifteenth month

The coefficient captures the difference insurvival probability between users initially en-rolled in a monthly contract and users initiallyenrolled in an annual contract The coefficientsin Table 6 are the marginal change in responseto an infinitesimal change in the continuousindependent variables and a discrete change forthe independent dummy variables In the spec-ification without controls (column 1) is pos-itive and significant Enrollment in a monthlycontract increases survival by 483 percentagepoints relative to the baseline rate of 3982percent survival with the annual contract Theintroduction of the controls increases the coef-ficient from 00483 to 00660 (column 2)Controlling for some of the unobserved hetero-

geneity reduces the downward bias on the co-efficient due to the initial sorting (Prediction 2)For example individuals enrolled with amonthly contract are significantly younger thanusers with an annual contract (Table 2) andyoung people are less likely to renew (column 2of Table 6) Failing to control for age biases thecoefficient downward

Finding 5 (survival probability) The survivalprobability after 14 months for the monthlycontract is 17 percent higher than for the an-nual contract

It is worth reiterating that ldquosurvivalrdquo includesrenewal with either of the two flat-rate con-tracts We can thus rule out that liquidity con-cerns (ie the difficulty of making an annualpayment all at once) and concerns about asecond long-term commitment for one year in-duce annual members to quit

RobustnessmdashIn columns 3 through 10 ofTable 6 we check the robustness of the findingsWe measure enrollment at the sixteenth month

TABLE 6mdashPROBIT OF RENEWAL DECISION

Sample Non-missing controls all clubs No subsidy all No subsidy II all

Dependent variableEnrollment at15th month

Enrollment at16th month




Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)

Dummy for enrollmentwith monthly contract

00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)

Corporate member 00728 00676 00676 0234 00024(00144) (00145) (00167) (00471) (00319)

Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)

Month and year ofenrollment X X X X X

Baseline renewalprobability for annualcontract 03983 04017 03906 03932 02609 02589 04701 05537 04252 04347

Number of observations N 4962 N 4962 N 4833 N 4833 N 2860 N 2860 N 715 N 715 N 1384 N 1384

Notes Standard errors in parentheses The number of observations is denoted by N Entries in the table represent the marginal coefficients of the probit in responseto an infinitesimal change in the continuous variables and a discrete change for the dummy variables An enrollment spell starts whenever an individual enrolls in theclub and ends whenever the individual quits or is censored The sample ldquonon-missing controlsrdquo consists of the individuals for whom the demographic controls ldquoagerdquoand ldquofemalerdquo are available The sample is further restricted to individuals who join at least 15 months before the end of the sample period The sample ldquono subsidyrdquois a restriction of the sample ldquonon-missing controlsrdquo to individuals paying on average a per-month fee of at least $70 The sample ldquono subsidy IIrdquo is a restriction ofthe sample ldquonon-missing controlsrdquo to individuals paying on average a per-month fee of at least $60 The controls ldquomonth and year of enrollmentrdquo indicate that the probitcontains 11 dummies for the month of enrollment and 4 dummies for year of enrollment The baseline renewal probability for the annual contract is the predictedrenewal probability for individuals starting with an annual contract

Significant at the 10-percent level Significant at the 5-percent level

Significant at the 1-percent level


after the joining date as an alternative measure ofsurvival With demographic controls users ini-tially enrolled in the monthly contract are 546percentage points more likely to be enrolled in thesixteenth month (column 4) than users initiallyenrolled in the annual contract Alternatively wemeasure enrollment at the twenty-seventh monthafter the joining date (columns 5 and 6) Theestimate of is positive although not significantlydifferent from zero

We also replicate the results of columns 1 and2 of Table 6 for the sample ldquono subsidyrdquo (col-umns 7 and 8) and for the larger sample ldquonosubsidy IIrdquo of users who pay at least $60 permonth in the monthly contract or $600 per yearin the annual contract (columns 9 and 10) In thefirst smaller sample the estimated has a sim-ilar magnitude as in the benchmark specifica-tion but the estimates are imprecise In thesecond wider sample the coefficient is positiveand large (01019 with controls) as well as pre-cisely estimated Overall the results on survivalprobability are robust to the measure of past at-tendance the measure of survival and the sample

Attendance over TimemdashFinally we test Pre-dictions 6 and 7 on the dynamics of averageattendance We first consider spells startingwith an annual contract in the sample ldquono sub-sidyrdquo and lasting at least two years22 We dis-play the results in columns 1 to 3 of the bottompart of Table 7

Finding 6 (average attendance over time inannual contract) In the annual contract av-erage monthly attendance for the initial groupin the first year 436 is significantly lower thanfor stayers in the second year 598

The difference in attendance between the twogroups is large the baseline group in the firstyear attends on average 27 percent less thanstayers in the second year Consequently theprice per average attendance in the first year$1522 is significantly higher than in the secondyear $1132 The results for the whole sample arecomparable (columns 4 to 6 of Table 7)

Figure 2A shows the within-year dynamics ofthe price per average attendance The sample at

month t is given by users in the ldquono subsidyrdquosample who have joined with an annual mem-bership and are still enrolled with a flat-ratecontract in the t-th month of tenure Over thefirst 12 months the price per average attendanceincreases from 123 to 19 as negative shocksaccumulate At renewal (months 13 and 14) theprice per attendance is halved

For spells starting with a monthly contractthe sample for average attendance at month t isgiven by the users in the ldquono subsidyrdquo samplewho have joined with a monthly membershipand are still enrolled with a flat-rate contract inthe t-th month of tenure Columns 1 to 3 of thetop part of Table 7 show the results by six-month groups

Finding 7 (average attendance over time inmonthly contract) Average monthly atten-dance in the first six months of a monthly con-tract 436 is 20 percent higher than in the nextsix months and is significantly higher than inany of the later six-month periods amongstayers

The price per average attendance in the firstsix months $1727 is significantly lower thanin any of the later six-month periods23 As Fig-ure 2B shows the price per average attendanceincreases over the first ten months from about$15 to about $20 and remains constant there-after The results are similar in the whole sam-ple (columns 4 to 6)

SummarymdashUnsubsidized monthly membersspend on average $187 for periods with noattendance before cancellation (Finding 4) de-spite small transaction costs of cancellation Inaddition after one year more monthly membersare still enrolled in a flat-rate contract thanannual members (Finding 5) Surprisinglymembers who pay higher fees for the option tocancel each month are more likely to renew pasta year This result does not arise because ofsorting but despite sorting (Finding 2) The re-sult is economically and statistically significantand robust across specifications Finally aver-age attendance decreases by 20 percent betweenthe first six months and the next six months in

22 The results remain unchanged if we restrict the samplefurther to users who renew with an annual contract after 12months

23 The results remain unchanged if we restrict the samplefurther to users who have had a monthly contract at all timesuntil month t


the monthly contract (Finding 7) a pattern op-posite to the one found for annual contracts(Finding 6)

IV Interpretations

We now consider which assumptions aboutconsumer preferences and beliefs can explainthe seven empirical findings summarized inTable 1 Two findings are consistent with stan-dard economic models Health clubs membersuse information on expected future attendanceto sort into the monthly and annual contract(Finding 2) and to sort out of the annual contract(Finding 6) The other findings instead are

hard to reconcile with the standard frameworkConsumers pay $17 per expected attendanceunder the monthly contract (Finding 1) and ap-pear to overestimate future attendance (Finding3) In addition monthly members with low at-tendance accumulate delays in cancellation(Finding 4) leading to a higher renewal proba-bility after one year relative to the annual con-tract (Finding 5) Finally average attendanceamong survivors decreases over time for themonthly contract (Finding 7) This finding ispuzzling since we observe the opposite patternfor the annual contract (Finding 6)

We first consider if enriched versions of thestandard model (Interpretations 1 and 2) can

TABLE 7mdashATTENDANCE AND PRICE PER AVERAGE ATTENDANCE OVER TIME

Sample No subsidy all clubs Sample All clubs


(1)


(2)

Average priceper averageattendance

(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)

N 866 N 866 N 866 N 6875 N 6875 N 6875Months 7ndash12 8189 363 2256 5281 391 1350

(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130

Users initially enrolled with an annual contract

Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108

Notes Standard errors in parentheses Standard errors for ldquoaverage price per average attendancerdquo measure computed usingthe bivariate Delta method The number of observations is denoted by N An enrollment spell starts whenever an individualenrolls in the club and ends whenever the individual quits or is censored The sample ldquono subsidyrdquo consists of the spells inwhich the average adjusted monthly fee is at least $70 if the spell starts with a monthly contract and at least $58 if the spellstarts with an annual contract For the six-month periods the sample includes spells that are ongoing not frozen and notmiscoded in at least one month in the period For year 1 in the annual contract the sample includes only spells that startedat least 14 months before the end of the sample period and that were not prematurely terminated because of medical reasonsor relocation For year 2 the sample includes only spells that started with an annual contract at least 26 months before theend of the sample period and that lasted at least 25 months The spells in row ldquofirst contract monthlyrdquo start with a monthlycontract The spells in row ldquofirst contract annualrdquo start with an annual contract The ldquoaverage pricerdquo in period t is the averagefee across people enrolled in period t The ldquoaverage attendancerdquo in period t is the average number of visits across peopleenrolled in period t The measure in column 3 is the ratio of the measure in column 1 and the measure in column 2


explain the additional findings We then discussnonstandard preferences and beliefs (Interpreta-tions 3 to 9) as possible explanations In the endwe summarize which explanations rationalizeall the empirical findings

1 Risk aversion Users who are risk aversein income may prefer a flat-rate contract to thepay-per-visit contract (Finding 1) because theformer contract minimizes the variance of the

payments24 Over the small amounts of moneyrequired for a monthly contract however

24 This result requires a utility function that is additivelyseparable in income and health club net benefits Under theassumption that the utility function is a concave function ofthe sum of income and health club net benefits the predic-tions are reversed more risk-averse agents are more likelyto choose the pay-per-visit contract

FIGURE 2 PRICE PER AVERAGE ATTENDANCE OVER TIME

Notes Point estimates and 95-percent confidence intervals plotted Figure 2A plots the ratioof average price and average attendance at month n of tenure The sample is ldquono subsidy allclubsrdquo for individuals initially enrolled in the annual contract and still enrolled at month n oftenure Figure 2B plots the ratio of average price and average attendance at month n of tenureThe sample is ldquono subsidy all clubsrdquo for individuals initially enrolled in the monthly contractand still enrolled at month n of tenure Standard errors for the ratio of average price andaverage attendance computed using the bivariate Delta method


health club members should be locally risk neu-tral (Matthew Rabin 2000)

2 Transaction costs Users may choose aflat-rate contract even though they attend little(Finding 1) if paying per visit entails largetransaction costs For the same reason they mayalso postpone the cancellation of a monthlycontract (Finding 4) However the actual trans-action costs are small Users can purchase aten-visit pass by filling out a simple form andcan then enter the club for ten visits with thesame procedure as users with a monthly orannual contract A transaction-cost-based expla-nation requires a time cost of over $70 for thefew minutes necessary to fill out the form Arelated explanation involves psychological trans-action costs such as distaste for payment per visit(Drazen Prelec and George Loewenstein 1998)These costs would also need to be high Moreoverthese explanations do not rationalize the over-estimation of future attendance (Finding 3) orthe differential renewal behavior for annual andmonthly contract (Findings 5 6 and 7)

3 Membership benefits Findings 1 and 4could arise from psychological benefits of themonthly and annual memberships These con-tracts may make the member feel ldquovirtuousrdquo orprovide the opportunity to impress others Ar-guably these psychological benefit should ap-ply also to ten-visit passes since in both casesconsumers complete an initial registration pro-cedure and receive a card which can be shownto friends Even if consumers treat monthly andannual memberships as special however it ishard to explain the differential renewal patternsfor monthly and annual contracts (Findings 5 to7) If anything the annual contract providesmore membership utility given that it signals astronger commitment This would imply ahigher survival probability for the annual con-tract against Finding 525

4 Time-variation in preferences for exer-cise If people enroll whenever they are mostenthusiastic about exercise a rational (but slow)updating process with mean reversion can ex-plain the delay in cancellation (Finding 4) andthe decrease in attendance among surviving

monthly members (Finding 7) Mean reversionhowever explains neither the initial overpay-ment (Finding 1) nor the difference betweenrenewal patterns of monthly and annual mem-bers (Findings 5 and 6)

5 Limited memory Rational agents withlimited memory may fail to cancel theirmonthly membership promptly after they stopattending (Finding 4) because they forget Dis-traction can also explain Findings 6 and 7 non-attenders fail to cancel in time but they getautomatically disenrolled under the annual con-tract Rational consumers however should an-ticipate their future limited memory and bewary of the monthly contract Instead over 90percent of customers with flat-rate contractschoose the monthly contract (Table 2) In addi-tion even if we allow for overestimation offuture memory this interpretation does not ex-plain Findings 1 and 3

6 Time inconsistency with sophisticationFlat-rate contracts are attractive to sophisticatedagents with ( ) preferences (Robert H Strotz1956 Edmund S Phelps and Robert A Pollak1968 Laibson 1997 Ted OrsquoDonoghue andRabin 1999) These agents have in addition tothe usual discount factor a discount factor 1 between present and future payoffs Theirdiscount function is 1 2 Given thathealth club attendance involves immediate costsand delayed benefits such present-biased agentsattend the health club less often than they wishat the time of enrollment They may purchase aflat-fee membership as a commitment devicethat increases future attendance (Finding 1)

These agents also delay one-time activitieswith immediate costs such as contract cancel-lation The cancellation delays of these agentsare too short however to account for Findings4 through 7 as we show with an extension ofthe calibrations in Section IIIA Using the samerevealed-preference argument we obtain abound on cancellation delay for sophisticatesgiven by E[T] max(4 [k2 k]s)26 Underthe calibrated magnitudes27 k2 $10 k 0 and

25 Taste for membership likely implies that high-attendance users switch from the monthly to the annualcontract to signal commitment This switch instead hap-pens for only 15 percent of the 6875 spells initiated witha monthly contract

26 The uniqueness of the equilibrium level of k can beproved along similar lines of Proposition 1 in James J Choiet al (2005)

27 Laibson et al (2004) M Daniele Paserman (2004)and Shui and Ausubel (2004) estimate the hyperbolic modelon field data and find values of between 05 and 08


08 nonattenders28 (s $233) delay atmost 533 days on average Under the sameassumptions low attenders (s $1) delay atmost 125 days These bounds do not depend onthe assumption 1 To show this we solvethe dynamic programming problem as a func-tion of assuming a discount factor 09995 (corresponding to a yearly discount fac-tor of 083) We consider the low-attendancecase (s $1) and assume k 13 N(15 4)29 Theresulting expected cancellation delay E[T] (Fig-ure 3A) is 5 days for 08 and is less than 15days even for a as low as 05 This calibrateddelay is substantially smaller than the observeddelay of over 60 days Figure 3B shows thecorresponding probability of a delay T of over120 days (4 months) This probability is essen-tially zero for all above 04 contrary to theempirical finding that 20 percent of users delayfor over 4 months Time inconsistency withsophistication therefore cannot generate thedelays observed in the data

7 Time inconsistency with partial naiveteAgents with ( )-preferences may be overcon-fident about their future self-control and expectto have a discount parameter with 1 (George A Akerlof 1991 OrsquoDonoghue andRabin 2001) These (partially) naive agentsmay pay more than $10 per expected visit(Finding 1) because they overestimate their fu-ture attendance (Finding 3) (This is in additionto the commitment device reason) We nowextend the calibrations in Section IIIA to showthat naive ( ) agents may also accumulatesubstantial delays in the cancellation of an au-tomatically renewed contract the other majorfinding in the paper Figure 3A plots the ex-pected cancellation delay for a naive agent withlow attendance (s $1) 09995 and costsk 13 N(15 4) For 07 the cancellationdelay of the naive agent matches the delay ofover 60 days observed in the data Moreoverthe same level of also matches the probabilityof delays lasting over 120 days (Figure 3B) 02Differently from time-consistent and time-inconsistent sophisticated agents the predicted

delay for naive agents matches the empiricalestimates A model of naive ( ) agents there-fore can explain all the findings in the paper30

8 Overestimation of net benefits Users maychoose flat-rate contracts (Finding 1) becausethey overestimate the future benefits of atten-dance or underestimate the expected futurecosts Projection bias (Loewenstein et al 2003)may reinforce the effect if health club consum-ers have high attendance expectations at sign-up This interpretation is consistent withFindings 3 and 4 but it does not explain Find-ings 5 6 and 8 on higher survival for themonthly than for the annual contract In orderfor overestimation to explain all of the empiricalfindings consumers need to have unrealisticexpectations about both the costs of attendanceand the costs of cancellation This is the case ifconsumers overestimate their future efficiencythat is their ability to perform desirable taskssuch as health club attendance and contractswitching

9 Persuasion Given that users attend onaverage fewer than eight times per month flat-rate contracts are on average more profitable forthe health clubs than pay-per-visit contractsHealth club employees therefore have incen-tives to persuade consumers to sign flat-ratecontracts They can do this either by not pro-viding (sufficient) information about the pay-per-visit alternative or by urging people to takeup the monthly or annual contract We addressthe first concern underprovision of informationby considering the contractual choices of a sub-group that is surely well-informed In our datamembers of a specific HMO can choose be-tween a 20-percent discount on the flat-ratecontracts and a $6 payment per visit Membersclaiming the discount must have obtained theinformation from the HMO itself which explic-itly lists both options Nevertheless the priceper expected attendance over months 1 to 6 forthe 1566 HMO members enrolling with amonthly contract equals $1031 (se 023) sig-nificantly higher than the $6 price per visitThus even informed members display the ten-dency to choose the more costly flat-ratecontract

28 The savings s for sophisticated agents include thebenefits of commitment to a higher future attendance underthe flat-rate contract (see DellaVigna and Malmendier2002)

29 The results are essentially insensitive to any choice of [10 30] and 2 [1 49]

30 The amount of delay predicted by the naive model isdecreasing in the variance of the cost distribution For substantially larger than 4 the calibrations of the naivemodel do not match the data


Alternatively health club employees mayexert pressure on members to choose a flat-rate contract (Finding 1) (B Douglas Bern-heim 1994) Employee persuasion may alsoexplain the cancellation lag for the monthlycontract (Finding 4) even though memberscan also cancel in writing Persuasion is un-

likely to explain the difference in renewalbetween the monthly and annual contract al-though health club employees can exert pres-sure to renew on both monthly members andannual members Persuasion does not explainthe survey evidence of overestimation of at-tendance (Finding 3)

FIGURE 3 CALIBRATION OF EXPECTED DELAY IN CANCELLATION


Out of the nine explanations above the mostsuccessful ones in our view involve both over-estimation of attendance and overestimation ofcancellation Overestimation of future atten-dance (Finding 3) leads consumers to chooseflat-rate contracts (Finding 1) Overestimationof future cancellation leads consumers to delaycancellation in the monthly contract (Finding4) but not in the annual contract which requiresno cost to cancel (Findings 5 to 7) A modelwith these features is the partially naive ( )model of OrsquoDonoghue and Rabin (2001) whichwe calibrate to the data A model of overesti-mation of future efficiency (which is not for-malized in the literature) would make the samepredictions without reference to self-control Inaddition persuasion by health club employeesis a plausible explanation for some of thefindings

A Heterogeneity

The leading explanation suggests that onemechanismmdashoverestimation of future self con-trol or of future efficiencymdashis at the root of allfindings If this is the case and there is hetero-geneity in overestimation we expect a correla-tion between the findings In particular monthlymembers who pay a high price per attendanceshould also be more likely to accumulate a longcancellation lag This is not necessarily the caseif the different findings are driven by differentphenomena (such as for example risk aversionfor Finding 1 and limited memory for Findings4 to 7)

We test this prediction for users enrolled inthe monthly contract As a measure of cancel-lation lags we use the number of consecutivefull months between the last attendance and theexpiration (as in Section IIIB) As a measure ofprice per attendance we take the ratio of thepayments to the health club over the attendancefor the period between sign-up and n monthsbefore the last attendance with n equal to 1 23 and 4 We limit the time frame in order toavoid a spurious correlation between the priceper attendance and months of delay due to lowattendance in the final months Finally we takethe log of 1 plus the measures in order to reducethe skewness of both variables The correlationbetween the cancellation lag and the price perattendance is positive and significant with val-ues between 0192 (n 1) and 0182 (n 4)

Longer lags n between the two measures do notaffect the estimate suggesting that the correla-tion is not likely to be spurious

Finding 8 (correlations) Users who pay a highprice per attendance in the monthly contractsubsequently display a longer gap between lastattendance and contract termination

These results are consistent with the idea thata unique explanationmdashsuch as overestimationof efficiency or self-controlmdashdrives both theresults on the high price per attendance forflat-rate memberships (Section IIB) and the re-sults on renewal behavior (Section IIIB)

V Conclusion

How do consumers choose from a menu ofcontracts In this paper we consider contractchoice in health clubs Using a new panel data-set from three US health clubs we find thatmembers who choose a contract with a flatmonthly fee of over $70 attend on averagefewer than 45 times per month They pay aprice per expected visit of more than $17 eventhough they could pay $10 per visit using aten-visit pass On average these users foregosavings of over $600 during their membershipWe also find that consumers who choose themonthly automatically renewed contract are 17percent more likely to stay enrolled beyond oneyear than users committing for a year This issurprising because monthly members payhigher fees for the option to cancel each monthWe present additional evidence including re-sults on cancellation delays and estimates ofattendance expectations from a survey Theseresults are difficult to reconcile with a standardmodel We present a number of explanations forthe findings The leading explanations involveoverestimation of future self-control or of futureefficiency

The analysis of consumer behavior is a firststep Rational profit-maximizing health clubscan observe the features of consumer behaviorusing datasets like the one analyzed in thispaper In DellaVigna and Malmendier (2004)we characterize the profit-maximizing contractfor goods with immediate costs and delayedbenefits such as health club attendance Forconsumers who are overconfident about futureself-controlmdashone of the leading explanations in


this papermdashthe profit-maximizing contract in-volves below-marginal cost pricing of atten-dance and automatic renewal with a transactioncost of cancellation The typical contract ofhealth clubs in the Boston area indeed has thesefeatures The evidence on contractual design isconsistent with the findings on consumerbehavior

DATA APPENDIX

The data on consumer behavior come fromthe attendance panel and the billing records Aseven-digit identification number allows us tolink multiple spells of the same individual

Attendance PanelmdashEach time a user with aflat-rate contract exercises a staff memberswipes the electronic card of the user and there-fore creates an attendance record An observa-tion of the attendance panel consists of theindividual identification number the date of thevisit basic demographic information (birthdaygender) a code for short-term membershipsand the enrollment and the expiration date (formembers who terminated the membership) Allinformation other than the date of visit is con-stant across the observations for a givenindividual

Billing RecordsmdashThe health clubs keep anofficial record of the customer payments Thebilling data provide detailed and accurate infor-mation about the category of usersmdashretail (thedefault) student family corporatemdashas well asthe type of transaction Each line of the billingpanel consists of the individual ID the date ofthe contractual transaction the four-digit codethat identifies the transaction and the price paid(if any) For example line ldquo1234567 1198R564 55rdquo indicates that user 1234567 paid anout-of-pocket monthly fee of $55 on January 11998 This monthly fee applies to employees ofthe company linked to code R564 For themonthly contract typical transactions are thepayment of the initiation fee the monthly feeand such items as an overnight locker or apersonal trainer Other codes involve monthlyfreezes of memberships bounced paymentsand termination of a membership for delin-quency in the payments For the annual con-tract typical transactions are the payment of theinitiation fee and of the annual fee

We use the price stated in the records as ameasure of the monetary payments to the clubsWe could alternatively use the four-digit codeand a conversion table (based on the prices as ofAugust 2000) to recover an imputed price Thecorrelation between the two measures of price is09668 None of the results changes if we usethe imputed price instead of the actual price

Monthly PanelmdashWe merge the attendanceand the billing panel into a unique dataset andwe then transform the data into a balanced panelwith monthly observations Each observationconsists of a variable defining the membership(not enrolledenrolled in a monthly contractenrolled in an annual contractin a freeze) thenumber of attendances in the month and theprice paid for the month For an annual contractthe monthly price is 1frasl12 the original price Weprorate the fees in the first month of monthlyand annual contracts that start in the middle ofa month We also prorate the fees in the finalmonth of an annual contract Monthly contractsalways terminate on the last day of the monthso no prorating is needed for the last month

Enrollment SpellsmdashWe define an enrollmentspell as the time period of continuous monthlyandor annual membership including possiblefreezes of the membership If no more than onefull calendar month of nonenrollment separatestwo contracts of an individual we still includethem in one spell For example this is the caseif an annual contract expiring on 11598 isrenewed on 31798 The missing monthly pay-ment may be due to an (unrecorded) one-monthpromotional offer a delay in payment or miss-ing data for a monthly payment

We consider an enrollment spell censored ifit is either ongoing at the end of the panel or ifit is followed by a short-term contract or apromotional membership Otherwise the spellis completed Short-term contracts are one-monthtwo-month three-month and four-month mem-berships with automatic expiration These areuncommon contracts designed for summer us-ers We identify promotional contracts as a se-quence of months with no contract andattendance in at least half of the months Weassume that in these periods health club mem-bers are using a free temporary membershipwhich the clubs grant in various promotional orcharitable initiatives


REFERENCES

Akerlof George A ldquoProcrastination and Obedi-encerdquo American Economic Review 199181(2) pp 1ndash19

Bernheim B Douglas ldquoA Theory of Confor-mityrdquo Journal of Political Economy 1994102(5) pp 841ndash77

Choi James J Laibson David Madrian Brigitteand Metrick Andrew ldquoOptimal Defaults andActive Decisionsrdquo National Bureau of Eco-nomic Research Inc NBER Working Pa-pers No 11074 2005

Cutler David M Glaeser Edward L and Sha-piro Jesse M ldquoWhy Have Americans Be-come More Obeserdquo Journal of EconomicPerspectives 2003 17(3) pp 93ndash118

DellaVigna Stefano and Malmendier UlrikeldquoContract Design and Self-Control Theoryand Evidencerdquo Quarterly Journal of Eco-nomics 2004 119(2) pp 353ndash402

DellaVigna Stefano and Malmendier UlrikeldquoOverestimating Self-Control Evidence fromthe Health Club Industryrdquo Stanford UniversityGraduate School of Business Research PaperNo 1880 2002

Eliaz Kfir and Spiegler Ran ldquoContracting withDiversely Naive Agentsrdquo Review of Eco-nomic Studies (forthcoming)

Ellison Glenn ldquoBounded Rationality in Indus-trial Organizationrdquo in Richard BlundellWhitney Newey and Tonsten Persson edsAdvances in economics and econometricsTheory and applications Ninth World Con-gress Cambridge Cambridge UniversityPress (forthcoming)

Fielding Helen Bridget Jonesrsquos diary A novelNew York Penguin 1999

Fielding Helen Bridget Jones The edge of rea-son New York Penguin 2001

Gabaix Xavier and Laibson David ldquoShroudedAttributes Consumer Myopia and Informa-tion Suppression in Competitive MarketsrdquoQuarterly Journal of Economics (forthcoming)

Gourville John T and Soman Dilip ldquoPaymentDepreciation The Behavioral Effects ofTemporally Separating Payments from Con-sumptionrdquo Journal of Consumer Research1998 25(2) pp 160ndash74

Heidhues Paul and Koszegi Botond ldquoThe Im-pact of Consumer Loss Aversion on PricingrdquoCenter for Economic Policy Research CEPRDiscussion Papers No 4849 2005

Hendel Igal and Nevo Aviv ldquoIntertemporal Sub-stitution and Storable Productsrdquo Journal ofthe European Economic Association 20042(2ndash3) pp 536ndash47

Laibson David ldquoGolden Eggs and HyperbolicDiscountingrdquo Quarterly Journal of Econom-ics 1997 112(2) pp 443ndash77

Laibson David I Repetto Andrea and Tobac-man Jeremy ldquoEstimating Discounted Func-tions from Lifecycle Consumption ChoicesrdquoUnpublished Paper 2004

Loewenstein George OrsquoDonoghue Ted andRabin Matthew ldquoProjection Bias in Predict-ing Future Utilityrdquo Quarterly Journal ofEconomics 2003 118(4) pp 1209ndash48

Madrian Brigitte C and Shea Dennis F ldquoThePower of Suggestion Inertia in 401(k) Par-ticipation and Savings Behaviorrdquo QuarterlyJournal of Economics 2001 116(4) pp1149ndash87

Miravete Eugenio J ldquoChoosing the Wrong Call-ing Plan Ignorance and Learningrdquo Ameri-can Economic Review 2003 93(1) pp 297ndash310

Miravete Eugenio J and Roller Lars-HendrikldquoCompetitive Non-Linear Pricing in DuopolyEquilibrium The Early US Cellular Tele-phone Industryrdquo Center for Economic PolicyResearch CEPR Discussion Papers No4069 2003

OrsquoDonoghue Ted and Rabin Matthew ldquoDoing ItNow or Laterrdquo American Economic Review1999 89(1) pp 103ndash24

OrsquoDonoghue Ted and Rabin Matthew ldquoChoiceand Procrastinationrdquo Quarterly Journal ofEconomics 2001 116(1) pp 121ndash60

Pagan Adrian and Ullah Aman Nonparametriceconometrics Cambridge Cambridge Uni-versity Press 1999

Paserman M Daniele ldquoJob Search and Hyper-bolic Discounting Structural Estimation andPolicy Evaluationrdquo Center for EconomicPolicy Research CEPR Discussion PapersNo 4396 2004

Phelps Edmund S and Pollak Robert A ldquoOnSecond-Best National Saving and Game-Equilibrium Growthrdquo Review of EconomicStudies 1968 35(2) pp 185ndash99

Prelec Drazen and Loewenstein George ldquoTheRed and the Black Mental Accounting ofSavings and Debtrdquo Marketing Science 199817(1) pp 4ndash28

Rabin Matthew ldquoRisk Aversion and Expected-


Utility Theory A Calibration TheoremrdquoEconometrica 2000 68(5) pp 1281ndash92

Shui Haiyan and Ausubel Lawrence M ldquoTimeInconsistency in the Credit Card MarketrdquoUnpublished Paper 2004

Strotz Robert H ldquoMyopia and Inconsistency inDynamic Utility Maximizationrdquo Review ofEconomic Studies 1956 23(3) pp 165ndash80

Tirole Jean The theory of industrial organiza-tion Cambridge MA MIT Press 1988


ior The dataset contains information both onthe type of membership and the day-to-day at-tendance decisions of 7752 health club mem-bers over three years We find that consumerschoose a contract that appears suboptimal giventheir attendance frequency In addition low-attendance consumers delay cancelling this con-tract despite small transaction costs

Our empirical analysis exploits the presenceof a contractual menu Consumers can choosebetween two flat-rate contractsmdasha monthlycontract and an annual contractmdashand a pay-per-visit option The monthly contract is automati-cally renewed from month to month until theconsumer cancels The annual contract insteadexpires after 12 months unless the consumerexplicitly renews it The variation in the per-usage pricing and in the renewal proceduresallows us to identify several puzzling features ofconsumer behavior

First consumers who choose a monthlymembership of over $70 per month pay onaverage 70 percent more than they would underthe pay-as-you-go contract for the same number

of visits Eighty percent of the monthly mem-bers would have been better off had they paidper visit for the same number of visits

Second consumers who choose the monthlycontract are 17 percent more likely to stay en-rolled beyond one year than users choosing theannual contract This is surprising becausemonthly members pay higher fees for the optionto cancel each month This result occurs eventhough high-attendance users sort into the an-nual contract at enrollment

These and additional empirical findings(summarized in Table 1) are hard to reconcilewith standard preferences and beliefs We ex-plore potential explanations including hightransaction costs of payment per usage riskaversion underestimation of costs of attendanceand of cancellation time inconsistency naiveteabout the time inconsistency and persuasion byhealth club employees

In our view the most parsimonious explana-tions are those allowing for overconfidence (na-ivete) Consumers overestimate for exampletheir future self-control or their future efficiency

TABLE 1mdashEMPIRICAL FEATURES AND POSSIBLE EXPLANATIONS

Standardmodel

Trans costsof paymentper usage

Membershipbenefits per

usageLimitedmemory

Timeinconsist

withsophistication

Time inconsistwith naivete

Overestimationof futureefficiency Persuasion

Finding 1Price per average attendance

$1727Distaste of

pay perusage

Membershipbenefits

Commitment Commitmentoverestimationof attendance


Pressure ofsalesman

Finding 2Average attendance in months

2ndash4 higher in annual thanmonthly contract









Finding 3Users predict 950 monthly

visits actual monthly visitsare 417



Finding 4Interval between last

attendance and termination231 full months

Distaste ofpay perusage

Membershipbenefits

Forget tocancel



Pressure ofsalesman

Finding 5Survival probability after 14

months 17 percent higherfor monthly than for annualcontract

Forget tocancel



Pressure ofsalesman

Finding 6Average attendance 27 percent

higher in second year forannual contract

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Learningsorting out

Finding 7Decreasing average attendance

over time in monthlycontract

Forget tocancel



Pressure ofsalesman

Finding 8Positive correlation of price

per average attendance andinterval between lastattendance and termination

Heterogeneity innaivete

Heterogeneity inoverconf












B Dataset





C Contractual Menu
































Allcontr

(1)

Allcontr

(2)

Allcontr

(3)

Allcontr

(4)


(5)

Firstcontractannual

(6)


(7)

Firstcontractannual

(8)


Total amount in $ 55830 55150 31408 51196 49840 61825 91802 102256(50052) (55150) (30418) (50052) (50494) (45071) (69958) (53689)


(2664) (1191) (1303) (2023) (1951) (2510) (4188) (4557)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204


(1857) (1909) (1097) (1922) (1919) (2058) (503) (1578)N 3185 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20

Annual contract 4819 4433 2413 4301 4699 4257 7012 6627(1564) (1708) (875) (1745) (1510) (1764) (454) (403)




(398) (408) (395) (401) (427) (396) (350) (395)N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204


(827) (903) (884) (866) (857) (208) (887) (226)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204



(094) (114) (072) (099) (104) (038) (120) (032)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Female 044 048 047 046 048 034 038 035(050) (050) (050) (050) (050) (047) (049) (048)


(844) (891) (930) (891) (878) (963) (975) (1086)N 3293 N 2745 N 1316 N 7354 N 6523 N 831 N 812 N 193



(021) (005) (005) (015) (015) (012) (005) (007)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204






(1)L

EF vaT p





EA v EM v















(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES












































B Dataset





C Contractual Menu
































Allcontr

(1)

Allcontr

(2)

Allcontr

(3)

Allcontr

(4)


(5)

Firstcontractannual

(6)


(7)

Firstcontractannual

(8)


Total amount in $ 55830 55150 31408 51196 49840 61825 91802 102256(50052) (55150) (30418) (50052) (50494) (45071) (69958) (53689)


(2664) (1191) (1303) (2023) (1951) (2510) (4188) (4557)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204


(1857) (1909) (1097) (1922) (1919) (2058) (503) (1578)N 3185 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20

Annual contract 4819 4433 2413 4301 4699 4257 7012 6627(1564) (1708) (875) (1745) (1510) (1764) (454) (403)




(398) (408) (395) (401) (427) (396) (350) (395)N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204


(827) (903) (884) (866) (857) (208) (887) (226)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204



(094) (114) (072) (099) (104) (038) (120) (032)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Female 044 048 047 046 048 034 038 035(050) (050) (050) (050) (050) (047) (049) (048)


(844) (891) (930) (891) (878) (963) (975) (1086)N 3293 N 2745 N 1316 N 7354 N 6523 N 831 N 812 N 193



(021) (005) (005) (015) (015) (012) (005) (007)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204






(1)L

EF vaT p





EA v EM v















(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES




































C Contractual Menu
































Allcontr

(1)

Allcontr

(2)

Allcontr

(3)

Allcontr

(4)


(5)

Firstcontractannual

(6)


(7)

Firstcontractannual

(8)


Total amount in $ 55830 55150 31408 51196 49840 61825 91802 102256(50052) (55150) (30418) (50052) (50494) (45071) (69958) (53689)


(2664) (1191) (1303) (2023) (1951) (2510) (4188) (4557)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204


(1857) (1909) (1097) (1922) (1919) (2058) (503) (1578)N 3185 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20

Annual contract 4819 4433 2413 4301 4699 4257 7012 6627(1564) (1708) (875) (1745) (1510) (1764) (454) (403)




(398) (408) (395) (401) (427) (396) (350) (395)N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204


(827) (903) (884) (866) (857) (208) (887) (226)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204



(094) (114) (072) (099) (104) (038) (120) (032)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Female 044 048 047 046 048 034 038 035(050) (050) (050) (050) (050) (047) (049) (048)


(844) (891) (930) (891) (878) (963) (975) (1086)N 3293 N 2745 N 1316 N 7354 N 6523 N 831 N 812 N 193



(021) (005) (005) (015) (015) (012) (005) (007)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204






(1)L

EF vaT p





EA v EM v















(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES





















































Allcontr

(1)

Allcontr

(2)

Allcontr

(3)

Allcontr

(4)


(5)

Firstcontractannual

(6)


(7)

Firstcontractannual

(8)


Total amount in $ 55830 55150 31408 51196 49840 61825 91802 102256(50052) (55150) (30418) (50052) (50494) (45071) (69958) (53689)


(2664) (1191) (1303) (2023) (1951) (2510) (4188) (4557)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204


(1857) (1909) (1097) (1922) (1919) (2058) (503) (1578)N 3185 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20

Annual contract 4819 4433 2413 4301 4699 4257 7012 6627(1564) (1708) (875) (1745) (1510) (1764) (454) (403)




(398) (408) (395) (401) (427) (396) (350) (395)N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204


(827) (903) (884) (866) (857) (208) (887) (226)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204



(094) (114) (072) (099) (104) (038) (120) (032)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Female 044 048 047 046 048 034 038 035(050) (050) (050) (050) (050) (047) (049) (048)


(844) (891) (930) (891) (878) (963) (975) (1086)N 3293 N 2745 N 1316 N 7354 N 6523 N 831 N 812 N 193



(021) (005) (005) (015) (015) (012) (005) (007)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204






(1)L

EF vaT p





EA v EM v















(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES







































Allcontr

(1)

Allcontr

(2)

Allcontr

(3)

Allcontr

(4)


(5)

Firstcontractannual

(6)


(7)

Firstcontractannual

(8)


Total amount in $ 55830 55150 31408 51196 49840 61825 91802 102256(50052) (55150) (30418) (50052) (50494) (45071) (69958) (53689)


(2664) (1191) (1303) (2023) (1951) (2510) (4188) (4557)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204


(1857) (1909) (1097) (1922) (1919) (2058) (503) (1578)N 3185 N 2551 N 1262 N 6951 N 6875 N 76 N 866 N 20

Annual contract 4819 4433 2413 4301 4699 4257 7012 6627(1564) (1708) (875) (1745) (1510) (1764) (454) (403)




(398) (408) (395) (401) (427) (396) (350) (395)N 436 N 391 N 147 N 974 N 97 N 877 N 6 N 204


(827) (903) (884) (866) (857) (208) (887) (226)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204



(094) (114) (072) (099) (104) (038) (120) (032)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204

Female 044 048 047 046 048 034 038 035(050) (050) (050) (050) (050) (047) (049) (048)


(844) (891) (930) (891) (878) (963) (975) (1086)N 3293 N 2745 N 1316 N 7354 N 6523 N 831 N 812 N 193



(021) (005) (005) (015) (015) (012) (005) (007)N 3495 N 2866 N 1391 N 7752 N 6875 N 877 N 866 N 204






(1)L

EF vaT p





EA v EM v















(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES





































(1)L

EF vaT p





EA v EM v















(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES








































(1)


(2)


(3)


Month 1 5523 345 1601(080) (013) (066)

N 829 N 829 N 829Month 2 8065 546 1476

(045) (019) (052)N 758 N 758 N 758

Month 3 7018 489 1434(105) (018) (058)

N 753 N 753 N 753Month 4 8179 457 1789

(026) (019) (075)N 728 N 728 N 728

Month 5 8193 442 1853(025) (019) (080)

N 701 N 701 N 701Month 6 8194 432 1895

(029) (019) (084)N 607 N 607 N 607

Months 1 to 6 7526 436 1727(027) (014) (054)

N 866 N 866 N 866


Year 1 6632 436 1522(037) (036) (125)

N 145 N 145 N 145














(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES













































(monthly fee $70)


(annual fee $700)


(1)

Price perattendance

(2)


(3)

Price perattendance

(4)


N 866 N 866 N 145 N 145




























ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES



























































ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES















































ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES












































ANNUAL CONTRACTS

(Sorting)


Sample All clubs




(01885) (01934) (01913)N 862 N 841 N 817

























































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES
























































































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES








































































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES































































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES















































Controls

Nocontrols

(1)

Controls time

dummies(2)

Nocontrols

(3)

Controls time

dummies(4)

Nocontrols

(5)

Controls time

dummies(6)

Nocontrols

(7)

Controls time

dummies(8)

Nocontrols

(9)

Controls time

dummies(10)


00483 0066 00337 00546 00011 00271 00634 00694 0091 01019(00218) (00221) (00221) (00224) (00260) (00254) (00479) (00501) (00368) (00372)

Female 00438 00425 00762 00187 00186(00143) (00144) (00165) (00394) (00277)

Age 00133 00155 00228 00304 00229(00046) (00046) (00052) (00111) (00077)

Age squared 00001 00002 00002 00003 00003(00001) (00001) (00001) (00001) (00001)


Student member 01123 00924 00894 01966 01173(00503) (00519) (00567) (02669) (00666)























IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES

















































IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES



































IV Interpretations







(1)


(2)


(3)


(4)


(5)


(6)


Months 1ndash6 7526 436 1727 4477 433 1035(027) (014) (054) (023) (005) (013)


(026) (017) (107) (031) (007) (026)N 577 N 577 N 577 N 3867 N 3867 N 3867

Months 13ndash18 8127 389 2088 5299 441 1203(034) (023) (126) (041) (010) (029)


(037) (031) (162) (059) (014) (039)N 189 N 189 N 189 N 1130 N 1130 N 1130


Year 1 6632 436 1522 4416 419 1055(037) (036) (125) (069) (016) (045)

N 145 N 145 N 145 N 598 N 598 N 598Year 2 6770 598 1132 4672 582 802

(107) (087) (167) (168) (045) (068)N 35 N 35 N 35 N 108 N 108 N 108




































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES



































































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES




























































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES
















































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES







































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES



































A Heterogeneity






V Conclusion





DATA APPENDIX









REFERENCES



































DATA APPENDIX









REFERENCES


































REFERENCES







































Date post:	05-Mar-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Paying Not to Go to the Gymulrike/Papers/gym.pdf · 2006-09-12 · Paying Not to Go to the Gym By S...

Documents