Job Search and Job Finding in a
Period of Mass Unemployment:
Evidence from High-Frequency
Longitudinal Data
by Alan Krueger and Andreas Mueller
Discussion by Bob Hall
NBER EF&G Meeting, New York Fed, February 4, 2011
·1
2
Search effort
Factor Relation to search effort Finding
SortingHigh effort searchers find jobs more quickly, leaving the low effort ones
Appears to be the opposite, but measurement problems may be the reason
Prior information
Searchers find out about available jobs fairly easily at the outset of search and then spend time waiting for prospects to materialize
Strongly supported
LearningSearchers reduce effort after early results are unfavorable
Appears to be strongly supported, but measurement problems may be the reason
WealthAs wealth is depleted, search becomes more intense
Appears to be strongly rejected but, measurement problems may be the reason
Unemployment insurance
Once benefits are exhausted, search becomes more intensive.
Not considered
3
Reservation wage
Factor Relation to reservation wage Finding
SortingLow reservation wage searchers depart unemployment soon, leaving high reservation wage searchers.
Rejected
Prior information
Wages are known at the outset, so there is no decline during unemployment
Strongly supported
LearningSearchers cut reservation wages after learning that higher-wage jobs are not available.
Rejected
WealthAs wealth is depleted, reservation wages decline
Rejected
Unemployment insurance
Benefits: Once exhausted, reservation wage declines.
Not discussed but implicitly supported
4
Old finding, confirmed here
42
Figure 2.1: UI weekly exit rate by UI duration
5
Two measures of search time
Time diary for the day before the survey: 7.6 hours per week
Recall question for the week: 11.5 hours per week
·
6
Two measures of search time
Time diary for the day before the survey: 7.6 hours per week
Recall question for the week: 11.5 hours per week
·
6
Minutes per day of search
43
Figure 3.1a: Time spent on job search (yesterday), in minutes per day
Figure 3.1b: Time spent on job search (last 7 days), in minutes per day
7
Related finding in the CPS
Bailar JASA, 1975
24 Journal of the American Statistical Association, March 1975
1. Rotation Group Indices in the CPS for Two Periods, 1968-69 (T1) and 1970-72 (T2), for Selected Characteristics
Monthly Std.
Characteristic Month in sample avg. error of Characteristic ~~~~~~~~~~~~~~~~~~~~~class size index
1 2 3 4 5 6 7 8 (000)
Total population 16 and over Civilian labor force Ti 102.3 100.3 99.8 99.5 100.8 99.3 99.1 99.0 80,340 0.3
T2 101.6 100.0 99.6 100.3 100.0 99.1 99.2 100.0 84,654 0.2
Employed Ti 101.6 100.2 99.9 99.8 100.4 99.4 99.4 99.3 77,285 0.3 T2 101.1 100.0 99.7 100.3 99.9 99.4 99.5 100.1 79,913 0.2
Unemployed Ti 120.0 101.5 96.4 92.8 109.3 96.5 92.6 91.0 3,055 2.4 T2 109.2 100.3 98.1 101.2 102.3 96.7 94.1 98.2 4,741 1.2
Hours worked per week 1-29 Ti 105.3 100.9 100.8 98.9 101.3 98.2 97.5 96.7 21,466 0.3
T2 103.9 101.1 99.8 100.1 100.7 98.4 97.3 98.7 35,560 0.2
30-34 Ti 101.1 101.0 99.1 100.0 98.2 100.1 100.1 100.6 8,894 0.8 T2 100.6 100.8 100.7 100.7 98.7 98.5 100.4 99.7 15,829 0.4
35-40 Ti 92.9 97.9 100.1 101.7 98.7 101.9 103.0 103.8 65,369 0.4 T2 93.1 97.7 99.9 101.5 99.2 101.7 103.1 103.9 103,797 0.2
41 or more Ti 113.0 103.5 99.3 96.8 102.6 96.2 94.9 93.6 42,972 0.2 T2 110.8 100.6 96.9 94.9 99.7 94.4 88.7 91.7 62,288 0.1
Males 16 and over Employed Ti 100.9 100.0 100.0 99.8 100.2 99.8 99.7 99.7 48,589 0.3
T2 100.7 99.9 99.8 100.2 99.9 99.7 99.7 100.2 49,637 0.2
Unemployed Ti 114.1 102.6 98.0 95.6 106.0 97.7 93.4 92.6 1,490 3.5 T2 105.4 101.4 99.9 101.6 100.3 98.0 95.6 97.9 2,578 1.6
Females 16 and over Civilian labor force TI 104.0 100.6 99.5 99.2 101.4 98.7 98.4 98.1 30,261 0.6
T2 102.7 100.0 99.4 100.4 100.2 98.6 98.6 100.0 32,439 0.5
Unemployed Ti 125.5 100.2 94.6 90.5 112.5 95.1 9.1.8 89.5 1,564 2.8 T2 113.8 99.0 95.9 100.7 104.4 95.2 92.2 98.6 2,163 1.5
estimate of the number of unemployed 20 percent higher than the average for all rotation groups; the rotation group in sample for the last time gives an estimate of the number of unemployed nine percent below the average. These estimates are very different and have different expected values.
The indices are shown for two time periods separately because of a change in procedure starting in 1970. As a result of the Gordon Committee Report [10] in 1962, it was decided to collect more information on those re- ported as not in the labor force. The questions were first included on the CPS questionnaire in 1967. The first time period shown in Table 1 is the two-year period 1968-69 when additional questions were asked of persons classified as not in the labor force and who were in sample for the first or fifth times. (The questions are shown in the exhibit as Questions 24A-E.) It was hypothesized that this procedure might account for a part of the rotation group bias on unemployment items because the inter- viewers might sometimes use the answers to these questions to reclassify these persons as in the labor force.
Therefore, in 1970 it was decided to ask these questions of persons in sample for the fourth and eighth times. The
second time period shown in Table 1 is the three-year period 1970-72. Changing the interview time at which these additional questions were asked did, in fact, reduce the indices for the rotation groups in sample for the first and fifth times and increase the indices for the rotation groups in sample for the fourth and eighth times. For the period 1970-72, the index for unemployment for the first month has been reduced to 109 and for the eighth month has been increased to 98. We conclude that the interview at which additional questions are asked of persons not in the labor force causes some of the difference in the estimates of the number of unemployed. The rotation group indices for unemployment are now more compar- able with those of earlier times.
The differences among estimates for the different rotation groups shown in Table 1 cannot be accounted for by sampling error; hence a rotation group bias must exist in the reporting, at least for some items. Another conclusion than can be drawn from the results in Table 1 is that the asking of probing questions of people not in the labor force changes the classification of some persons and hence causes a difference in the number of persons classified in certain categories.
8
Regressions of search time on
unemployment duration
Table 3.1a Linear regressions of time spent on job search (yesterday), with and without fixed effects
Dependent varialbe:
time spent on job search, in mins. per day Week 1
Pooled
cross-
section
Fixed
effects
Fixed
effects
Unemployment duration, in weeks 0.227 -0.075 -2.73 -1.62
(0.104)** (0.072) (0.250)*** (0.313)***
Lapse (before November 8) -0.937
(6.924)
Exhausted UI 8.416
(10.724)
After extension of November 8 -19.056
(3.039)***
Log(weekly benefit amount) -19.303 -18.78
(12.570) (10.311)*
Log(weekly previous wage) 10.781 14.554
(7.284) (6.113)**
Controlling for age, education, sex, race and ethnicity x x
Dummies for day of week of diary x x x x
Individual fixed effects x x
Mean of dependent variable 103.1 65.5 65.1 65.1
Min 0 0 0 0
Max 720 780 780 780
N 3,924 24,366 25,180 25,180
R-squared 0.10 0.05 0.66 0.67
Robust standard errors in parentheses (clustered at individual level); * p<0.1, ** p<0.05, *** p<0.01.
Notes: Survey weights are used. Universe: Unemployed; no job offer yet accepted; age 20-65.
Table 3.1b Linear regressions of time spent on job search (last 7 days), with and without fixed effects
Dependent varialbe:
time spent on job search, in mins. per day Week 1
Pooled
cross-
section
Fixed
effects
Fixed
effects
Unemployment duration, in weeks 0.093 0.059 -2.245 -1.538
(0.130) (0.134) (0.288)*** (0.331)***
Lapse (before November 8) -1.096
(7.994)
Exhausted UI -1.415
(13.978)
After extension of November 8 -11.788
(3.651)***
Log(weekly benefit amount) -2.921 -28.141
(16.950) (15.924)*
Log(weekly previous wage) 18.855 38.659
(10.931)* (10.370)***
Controlling for age, education, sex, race and ethnicity x x
Individual fixed effects x x
Mean of dependent variable 117.5 98.3 97.6 97.6
Min 0 0 0 0
Max 685.7 685.7 685.7 685.7
N 3,983 24,638 25,449 25,449
R-squared 0.04 0.05 0.77 0.77
Robust standard errors in parentheses (clustered at individual level); * p<0.1, ** p<0.05, *** p<0.01.
Notes: Survey weights are used. Universe: Unemployed; no job offer yet accepted; age 20-65.
9
Marginal probit coefficients for
probability of early UI exit
Table 5.2 Probit models (marginal effects) for leaving UI early and receiving a job offer
Dependent Variable:
Left UI early
(before March 14, 2010) Received job offer
Explanatory Variables: (1) (2) (3) (4) (5) (6)
Time spent on job search, in hours per week 0.0018 0.0018 0.0017 -0.0002 -0.0002 -0.0002 (0.0006)*** (0.0005)*** (0.0005)*** (0.0007) (0.0006) (0.0006)
Log(reservation wage ratio) -0.0492 -0.0485 -0.0517 (0.0255)* (0.0246)** (0.0252)**
Cohort 2 0.05 0.0284 0.0372 0.0388 (0.0352) (0.0331) (0.0375) (0.0386)
Cohort 3 -0.0122 -0.0155 -0.0181 -0.0112 (0.0277) (0.0271) (0.0301) (0.0303)
Cohort 4 -0.0352 -0.0393 -0.0119 -0.0135 (0.0276) (0.0275) (0.0313) (0.0312)
Cohort 5 -0.0966 -0.0912 0.0343 0.0394 (0.0202)*** (0.0202)*** (0.0371) (0.0371)
Cohort 6 -0.0593 -0.0544 0.0015 -0.0044 (0.0285)** (0.0287)* (0.0373) (0.0358)
Cohort 7 -0.0991 -0.0958 -0.0795 -0.0767 (0.0204)*** (0.0204)*** (0.0227)*** (0.0240)***
Cohort 8 -0.0872 -0.0784 -0.0189 -0.0078 (0.0196)*** (0.0207)*** (0.0283) (0.0300)
Recall expectation 0.0003 0.0508 (0.0322) (0.0407)
With recall date 0.3872 0.0633 (0.1533)** (0.1196)
Age 0.0012 0.003 (0.0047) (0.0052)
Age^2 0.000 0.000 (0.0001) (0.0001)
High school degree 0.0025 0.0312 (0.0538) (0.0639)
Some college education 0.0343 0.0799 (0.0540) (0.0633)
College degree 0.1018 0.1185 (0.0652) (0.0738)
Some graduate education 0.0954 0.2085 (0.0792) (0.1005)**
Graduate degree 0.1356 0.0928 (0.0773)* (0.0776)
Black -0.0055 0.0062 (0.0241) (0.0263)
Asian or other 0.016 0.0007 (0.0356) (0.0362)
Race not available 0.0176 0.0786 (0.0354) (0.0473)*
Hispanic -0.0106 0.0195 (0.0306) (0.0381)
Ethnicity not available -0.0037 -0.0831 (0.0305) (0.0238)***
Female 0.0109 -0.0051 (0.0180) (0.0187)
Married 0.0363 0.0092 (0.0199)* (0.0231)
Mean of dependent variable 0.1653 0.1653 0.1649 0.1611 0.1611 0.1614
Min 0 0 0 0 0 0
Max 1 1 1 1 1 1
N 4,433 4,433 4,382 3,918 3,918 3,864
Pseudo R-squared 0.0084 0.0348 0.0589 0.000 0.0077 0.0313
Robust standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01. Notes: Survey weights are used. Sample: unemployed; no job offer yet accepted. For the job offer regressions, only respondents with two or more interviews were
included.
10
Identification
Search productivity: hi = αi + si
Exit benefit: hi −1
2h2
i
Search time cost: γisi +1
2s2
i
·
11
Identification
Search productivity: hi = αi + si
Exit benefit: hi −1
2h2
i
Search time cost: γisi +1
2s2
i
·
11
Identification
Search productivity: hi = αi + si
Exit benefit: hi −1
2h2
i
Search time cost: γisi +1
2s2
i
·
11
Identification, continued
maxsi
αi + si −1
2(αi + si)
2 − γisi −1
2s2
i
FONC: 1− hi − γi − si = 0
·
12
Identification, continued
maxsi
αi + si −1
2(αi + si)
2 − γisi −1
2s2
i
FONC: 1− hi − γi − si = 0
·
12
Two-equation system
Search productivity: hi = αi + si
Optimal time allocation to search: hi = 1− γi − si
·
13
Two-equation system
Search productivity: hi = αi + si
Optimal time allocation to search: hi = 1− γi − si
·
13
Two dimensions of heterogeneity
3 5
4
High productivity
3
3.5 High productivity
2.5
rd
Low productivity
2
Exit haza
1
1.5 Low search time cost
0.5 High search time cost
0
0 0.5 1 1.5 2 2.5 3Search time
cost
14
What the econometrician sees
3 5
4
3
3.5
2.5
rd
2
Exit haza
1
1.5
0.5
0
0 0.5 1 1.5 2 2.5 3Search time
15
Wage-setting typology from
Hall-KruegerModels of Wage Formation
Commitment to ignore counteroffers
Wage offer customized to worker
Interruption to alternating offer bargaining likely?
Commitment to ignore counteroffers
Wage offer customized to worker
Diamond paradox
Posted wage
Wage tightly linked to conditions
Wage less responsive
to conditions
Interruption to alternating offer bargaining likely?
3 16
Reservation wage and
unemployment duration
Table 4.1 Reservation wage ratio by duration of unemployment
All
durations
Less
than 5
weeks
5 - 9
weeks
10 - 14
weeks
15 - 19
weeks
20 - 24
weeks
25 - 49
weeks
50 +
weeks
Feldstein & Poterba (1984):
All Job Losers and Leavers 1.07 1.11 1.09 1.04 1.06 1.04 1.02 0.99
Feldstein & Poterba (1984):
Job Losers 1.03 1.06 1.05 1.03 1.06 1.00 0.99 0.97
Krueger & Mueller:
Cross-section (1st week) 0.99 1.04 1.02 1.01 1.00 1.06 0.95 0.94
Krueger & Mueller:
Longitudinal estimate 0.99 1.00 1.00 1.00 0.99 0.99 0.98 0.97
Note: Survey weights are used. Universe: Unemployed; no job offer yet accepted; age 20-65.
Feldstein and Poterba's (1984) estimates are from a sample of 2,228 unemployed from the May 1976 Current Population Survey.
17
Probability of acceptance when
there is a threshold at W = R
0.9
1.0
0 7
0.8
ty
0.6
0.7
prob
abili
0.4
0.5
eptance p
0.2
0.3Acce
0.0
0.1
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4
Ratio of wage offer to reservation wage
18
W ≥ R matters
Table 6.1a Hourly offered wage below and above hourly reservation wage
Hourly offered wage <
hourly reservation wage
Hourly offered wage >=
hourly reservation wage
Accepted 50.5% 74.1%
Not accepted 23.6% 10.3%
Undecided 25.9% 15.7%
N 566 587
Table 6.1b Hourly offered wage below and above hourly reservation wage (full time offers only)
Hourly offered wage <
hourly reservation wage
Hourly offered wage >=
hourly reservation wage
Accepted 44.4% 73.8%
Not accepted 24.2% 11.4%
Undecided 31.4% 14.8%
N 361 417
19
W ≥ R matters somewhere near 0
Table 6.2 Marginal effects from probit model for accepting a job offer, conditional on receiving an offer
(1) (2) (3) (4)
Hourly offered wage ≥ Hourly reservation wage (lagged) 0.212 0.196 0.206 (0.098)** (0.061)*** (0.095)**
Hourly offered wage ≥ Hourly reservation wage (lagged)
* Part time job offer -0.215 -0.209 (0.127)* (0.126)*
Hourly offered wage ≥ Hourly previous wage 0.152 0.106 (0.101) (0.070)
Hourly offered wage ≥ Hourly previous wage
* Parttime job offer -0.055 (0.126)
Parttime job offer 0.21 0.152 0.207 (0.079)*** (0.079)* (0.080)***
Log(lagged hourly reservation wage) -0.109 -0.104
(0.122) (0.127)
Log(hourly offered wage) 0.216 0.185 0.222 (0.108)** (0.091)** (0.107)**
Log(hourly previous wage) -0.029 (0.098)
Savings < $10,000 0.013
(0.088)
Unemployment duration, in weeks 0.001
(0.001)
Mean of dependent variable 0.61 0.61 0.61 0.61
Min 0 0 0 0
Max 1 1 1 1
N 1,153 1,153 1,153 1,153
Pseudo R-squared 0.07 0.05 0.05 0.07
Robust standard errors in parentheses (clustered at individual level); * p<0.1, ** p<0.05, *** p<0.01.
Notes: Survey weights are used. Sample: Respondents age 20-65.
20
Discontinuity in probit
coefficient
48
Figure 6.1: Effect of Alternative Cutoffs for Reservation Wage Relative to Offered Wage
on Chance of Accepting Job
Notes: Figure shows the effect of varying the reservation wage threshold on job acceptance.
Specifically, if the reservation wage is R and the offered wage is W, a binary variable was
created that equaled one if W ≥ (1+X)R, and zero otherwise. The horizontal axis shows the
effect of alternative values of X on job offer acceptance.
21
Offers are rare
47
Figure 5.1: Cumulative Probability of Receiving at Least One Job Offer
22
And most are accepted
61 percent
Hardly any job-seekers see more than one offer
This puts a lot of tension on job-seekers’ beliefs about theoffer distribution—they don’t learn much about thedistribution while searching
One-armed bandit model not relevant
·
23
And most are accepted
61 percent
Hardly any job-seekers see more than one offer
This puts a lot of tension on job-seekers’ beliefs about theoffer distribution—they don’t learn much about thedistribution while searching
One-armed bandit model not relevant
·
23
And most are accepted
61 percent
Hardly any job-seekers see more than one offer
This puts a lot of tension on job-seekers’ beliefs about theoffer distribution—they don’t learn much about thedistribution while searching
One-armed bandit model not relevant
·
23
And most are accepted
61 percent
Hardly any job-seekers see more than one offer
This puts a lot of tension on job-seekers’ beliefs about theoffer distribution—they don’t learn much about thedistribution while searching
One-armed bandit model not relevant
·
23
Search effort
Factor Relation to search effort Finding
SortingHigh effort searchers find jobs more quickly, leaving the low effort ones
Appears to be the opposite, but measurement problems may be the reason
Prior information
Searchers find out about available jobs fairly easily at the outset of search and then spend time waiting for prospects to materialize
Strongly supported
LearningSearchers reduce effort after early results are unfavorable
Appears to be strongly supported, but measurement problems may be the reason
WealthAs wealth is depleted, search becomes more intense
Appears to be strongly rejected but, measurement problems may be the reason
Unemployment insurance
Once benefits are exhausted, search becomes more intensive.
Not considered
24
Reservation wage
Factor Relation to reservation wage Finding
SortingLow reservation wage searchers depart unemployment soon, leaving high reservation wage searchers.
Rejected
Prior information
Wages are known at the outset, so there is no decline during unemployment
Strongly supported
LearningSearchers cut reservation wages after learning that higher-wage jobs are not available.
Rejected
WealthAs wealth is depleted, reservation wages decline
Rejected
Unemployment insurance
Benefits: Once exhausted, reservation wage declines.
Not discussed but implicitly supported
25