+ All Categories
Home > Documents > Stated Preferences and Intrahousehold Resource Allocation: Evidence from Northern Ethiopia

Stated Preferences and Intrahousehold Resource Allocation: Evidence from Northern Ethiopia

Date post: 29-Nov-2023
Category:
Upload: independent
View: 0 times
Download: 0 times
Share this document with a friend
32
Stated Preferences and Intrahousehold Resource Allocation: Evidence from Northern Ethiopia DRAFT 08/06/98 Julian A. Lampietti Abstract This paper reports on a new test of the restrictions implied by the unitary model of household behavior. We ask a sample of married men and women in Northern Ethiopia contingent valuation questions about how they would allocate preventive health care across members of their households. Our results indicate that married men’s and women’s stated preferences for malaria prevention can not be pooled and be treated as identical in the context of the unitary model. This provides new evidence against the unitary model of household behavior. * This investigation received financial support from the UNDP/World Bank/WHO Special Programme for Research and Training in Tropical Diseases (TDR) = Please address correspondence to: Julian Lampietti, 1613 13th Street, N.W., Washington, D.C. 20009. Tel: 202 518 9258 Email: [email protected] or [email protected]
Transcript

Stated Preferences and Intrahousehold Resource Allocation:

Evidence from Northern Ethiopia

DRAFT

08/06/98

Julian A. Lampietti

Abstract

This paper reports on a new test of the restrictions implied by the unitary model of household behavior. We ask asample of married men and women in Northern Ethiopia contingent valuation questions about how they wouldallocate preventive health care across members of their households. Our results indicate that married men’s andwomen’s stated preferences for malaria prevention can not be pooled and be treated as identical in the context ofthe unitary model. This provides new evidence against the unitary model of household behavior.

* This investigation received financial support from the UNDP/World Bank/WHO Special Programme for Research and Training inTropical Diseases (TDR)= Please address correspondence to: Julian Lampietti, 1613 13th Street, N.W., Washington, D.C. 20009.Tel: 202 518 9258 Email: [email protected] or [email protected]

TABLE OF CONTENTS

I. INTRODUCTION ...............................................................................................................................3

II. BACKGROUND................................................................................................................................5

PREVIOUS TESTS OF THE UNITARY MODEL ......................................................................................................5CONTINGENT VALUATION.............................................................................................................................6

III. SURVEY INSTRUMENT AND RESEARCH DESIGN .................................................................9

SURVEY INSTRUMENT...................................................................................................................................9RESEARCH DESIGN......................................................................................................................................9

IV. STUDY SAMPLE........................................................................................................................... 11

KNOWLEDGE OF MALARIA.......................................................................................................................... 11MALARIA INCIDENCE ................................................................................................................................. 11COST OF ILLNESS ...................................................................................................................................... 12SOCIOECONOMIC CHARACTERISTICS ........................................................................................................... 12

V. ECONOMETRIC MODEL ............................................................................................................. 14

VI. RESULTS ....................................................................................................................................... 16

VALIDATION OF RESULTS............................................................................................................................ 16STATED PREFERENCES............................................................................................................................... 16

VII. CONCLUSIONS ........................................................................................................................... 18

REFERENCES ..................................................................................................................................... 19

I. Introduction 1

It has long been recognized that intrahousehold resource allocation decisions can have a profound effect

on the schooling, health, and nutrition of individual household members. The choice of a household model has

important implications for policy formulation and implementation.2 For example, policy makers have often

suggested that increasing women’s access to income can be especially beneficial for the health of children. At the

same, they have tended to rely on a “unitary” model that assumes individual household members have identical

preferences or a single decision maker’s preferences hold for the entire household. This model implies that

household expenditures are independent of which individuals in the family receive income or control household

resources.

Dissatisfaction with these implications led to the development of “collective” models that incorporate

divergent preferences of individual household members. The early 1980’s saw the development of cooperative

bargaining models where individuals have a choice of remaining single or forming a household (Manser and

Brown, 1980; McElroy and Horney, 1981; and Lundberg and Pollak, 1983). This was followed by the

development of non-cooperative bargaining models where individual choices are conditional on the choices of

other household members (Kanbur and Haddad, 1994; and Lundberg and Pollak, 1994). Chiappori (1988, 1992)

proposed a flexible collective model which assumes only that the outcome of a cooperative household decision

process is Pareto efficient.

A household behavioral model should be used to develop policy only after the restrictions it implies can

not be rejected by the data. To date, empirical testing has focused almost exclusively on the unitary model. Despite

evidence against it, there has been no conclusive rejection of this model. This is because previous tests have relied

on a measure of husband’s and wife’s relative control over household resources, including earned income,

unearned income, and income transfers, that are not exogenous to past or present household behavior (Lundberg

and Pollak, 1996).

In this paper we take a completely new approach to testing the unitary model. We ask a sample of married

men and women contingent valuation questions about how they would allocate preventive health care across

members of their households. Our test rests on the assumption that all married men have comparable utility

functions and that all married women have comparable utility functions. The logic is as follows: if individuals in

households have identical preferences, as implied by the unitary model, then there should not be a significant

1 The authors would like to thank the following individuals for their contributions to this research: C. Poulos, D.

Whittington, M. Cropper, V.K. Smith, H. Mitiku, K. Komives, T. Ghebreyesus, and K. Witten.

2 There are several excellent reviews of the literature on models for intrahousehold resource allocation, including Behrman(1997), Haddad et al (1997), Strauss and Beegle (1996), Strauss and Thomas (1995), and Alderman et al (1995).

difference in men’s and women’s stated preferences for the allocation of preventive health care in their households.

To the extent that there is a difference in stated preferences, this is a violation of the unitary model of household

behavior. An additional advantage of this test is that the choice offered to the respondents is completely exogenous

to past and present household behavior.

Let us be clear that this is not a test of the household decision making process or the outcome of this

process. It is a test of the maintained assumptions of the unitary model of household behavior - that household

decision makers have identical preferences. We test our hypothesis using a survey of married men and women

from randomly selected households in the Tigray region of Northern Ethiopia. Our results indicate that married

men’s and women’s stated preferences for malaria prevention can not be pooled and be treated as identical in the

context of the unitary model, providing new evidence against the unitary model of household behavior.

The inference we draw from our reserach design depends on the equivalence of households in which

married men and women were interviewed. While the choice of households was random, there are two important

caveats to the design. First only one individual, either a married man or a married woman, was interviewed in each

randomly selected household. Second, there was no protocol for selecting whether a man or woman was

interviewed in each household. This opens the possibility for a systematic and unobserved difference in the

households in which men and women were interviewed. If this is the case, then our results may only be interpreted

as suggestive.

This remainder of this paper has 6 sections. Section II reviews previous tests of the unitary model and the

debate surrounding contingent valuation. Section III outlines the survey instrument and research design. Section

IV presents summary statistics for the study sample and section V develops an econometric model for the choice

question. Univariate and multivariate results are presented in section VI and conclusions in section VII.

II. Background

Empirical testing of the restrictions implied by intrahousehold models has focused almost exclusively on

the unitary model. This model implies that all income sources within the household are pooled. Thus a prediction

of the model is that only total household income should matter for resource allocation decisions, not who in the

household controls the income. This prediction is not always consistent with observation. For example, in many

developing countries men are observed spending income on goods for personal consumption such as alcohol and

tobacco (Phipps and Burton, 1992; Hoddinott and Haddad, 1995). In the same context, women are observed

purchasing goods for household consumption such as food and health care.

Previous Tests of the unitary model Previous Tests of the unitary model”\L2

Two approaches have been used to test the pooling hypothesis. The first compares earned income

expenditures on goods across family members. The second tests for differences in expenditure patterns from

unearned income and transfer income. The assumption in both cases is that if expenditure patterns from earned,

unearned, or transfer income differ consistently between husbands and wives, then this is evidence against pooling.

The empirical tests of pooling have shown that the earned income controlled by husbands and wives has a

significant and often different effect on family behavior, whether measured by expenditure on categories of goods

and services, or by outcomes such as child health. For example, increases in the wife’s income relative to the

husband’s income have been shown to be associated with greater expenditures on restaurant meals, child care and

women’s clothing (Phipps and Burton, 1992) and with reduced expenditures on alcohol and tobacco (Phipps and

Burton, 1992; Hoddinott and Haddad, 1995). Increases in child health, nutrition and survival probabilities have

also been associated with a mother’s control over family resources (Shultz 1990; Thomas 1990, 1993).

It is widely recognized that the interpretation of the separate effects of husband’s and wife’s earnings does

not represent conclusive evidence against the unitary model. Consider the following example: if a woman’s wages

were increased, economic theory predicts that she would purchase more preventive health care in order to change

her allocation of time away from caring for sick household members. Such a prediction is consistent with both the

unitary model and the collective model. In the former, the allocation of labor for caring for the sick is reorganized

to increase the women’s labor force participation. In the latter, the woman has gained more control over household

resources and the increased expenditures on preventive health care reflect her preferences.

In order to circumvent this problem, non-earned income has been offered as an appropriate exogenous

measure of resource control. Shultz (1990) was the first to note that “if non-earned income (or ownership of an

underlying asset ) influences family demand behavior differently, depending on who in the family controls the

income (or owns the asset), then the preferences for that demand must differ across individuals and such families

must not completely pool non-earned income.” Thomas (1990) examines the demand for per capita caloric and

protein intakes, fertility, child survival, and weight-for-height for children and rejects income pooling using non-

labor income in the hands of men and women. Thomas (1993) finds that income in the hands of women is

associated with a larger increase in the share of the household budget devoted to household services, health and

education. Although non-earned income is not contaminated by price effects, it does not avoid the problems of

interpretation alluded to in the previous paragraph because it is not completely exogenous to past or present

household behavior (Lundberg and Pollak, 1996).

Lundberg, Pollak and Wales (1997) use transfer income to reject the pooling hypothesis. They find that

expenditures on women’s and children’s goods increased coincident with a policy change that transferred a

substantial child allowance from husbands to wives in the United Kingdom. There are two problems with their

test. First, their research design depends on pre-and post-policy comparison of treatment groups.3 If households

dropped out of the survey in a way that is systematically correlated with the change in policy, and this is not

accounted for in the analysis, then their inference may break down. There appears to be just such attrition in the

data: the number of observations in their study decreases from 181 in the first time period to 118 in the second and

to 82 in the third. Second, as the authors rightly point out, although their findings are consistent with the

proposition that the policy change caused the change in consumption patterns, they do not prove causation. Indeed,

the shift in consumption patterns that they attribute to the change in resource control may be due to changes in

preferences for children’s clothing or changes in relative prices.

The unitary model has been criticized for not standing up well to empirical testing. None of these tests,

however, has proved to be a conclusive rejection of the pooling hypothesis. An ideal test this hypothesis is an

experiment in which randomly selected husbands and wives in the same household are separately offered the

opportunity to purchase a good that provides benefits to members of their household. If the unitary model holds,

then we expect husbands and wives to have make the same choices.

Contingent valuation

This paper relies on contingent valuation (CV) questions to test the maintained assumptions of the unitary

model. CV is an increasingly popular technique for informing public policy decisions. The range of applications

includes determining compensation for environmental damages as well as the identification and design of public

sector projects. In both cases, the purpose of a CV survey is to provide decision makers with information about a

population’s willingness to pay (WTP) for a good or service. Despite increasing popularity, there remains

3 Their data (the United Kingdom Family Expenditure Survey) is based on an annual random cross section of households.

considerable skepticism in the literature as to whether responses to CV questions produce accurate measures of

WTP.

A CV question consists of framing a scenario and then asking the respondent to answer a question about a

change in the level of provision of a good or service. CV practitioners assume their subjects have the same

incentives to reveal real economic commitments when faced with hypothetical choices as they do in real markets.

There is, however, no a priori basis for this assumption. In a real market, the individual’s incentive to reveal the

truth is that their consumption choice is constrained by a fixed budget: a commitment to pay for a good implies a

conscious sacrifice in the acquisition of other goods. Solow and Arrow (1993), two nobel prize winners, identify

the core of the incentives issue by asking the question: “do people make the same kind of choices when asked what

they might pay for something as they do when they are faced with circumstances where they must pay to acquire

something?”

The validity issue in CV is complicated by the general absence of a measurable ‘true’ WTP value which

can be used to assess the bias of responses. Validation is necessary because there are many factors that potentially

make responses to CV questions unreliable or invalid. Unreliable responses might be given because the respondent

is unfamiliar with the good or service being offered, or has not taken the survey seriously. Invalid responses occur

when the CV question does not accurately measure the respondent’s WTP, such as when WTP is overstated or

understated in order to influence the outcome of the survey.

This view is amplified in the literature, where there is a lack of consensus concerning the reliability and

validity of WTP responses to CV surveys. There are 17 studies that compare stated WTP with “real” or

behavioral WTP (e.g. WTP for goods that can actually be bought and sold). Nine of these studies compare stated

WTP with actual WTP in a field setting. The design and methods used in most of these have been criticized by

both Smith (1993) and Haneman (1994). The findings were also reviewed by Solow and Arrow (1993) with the

tentative conclusion that stated WTP may be greater than actual WTP. Eight of these studies rely on experimental

economics settings. Although there are many advantages to experimental approaches, there is no conclusive

evidence to suggest that these results can be applied to a real world setting. Much concern exists about the

external validity or possible lack of correspondence between results obtained in the field and results obtained in the

laboratory. Solow and Arrow (1993) argue that one cannot readily justify drawing inferences from subject

behavior in experimental economics to subject behavior in CV on the basis that the payment mechanisms used in

experimental economics are quite different from those used in CV.

A recent literature review by Carson et al (1996) indicates there are 83 studies that specifically try to

establish a correspondence between stated WTP and revealed preference. Carson et al conclude that “CV

estimates are smaller, but not grossly smaller, than their revealed preference counterparts.” Smith (1993) reviews

selected revealed preference studies and concludes that “for the most part, they support CV estimates as being

‘comparable’ in performance to the alternative approach providing the reference point (or standard) in each case.

In some cases the two estimates might be judged significantly different but still exhibit a strong causal

relationship.”

In our survey we use multiple strategies to promote and test for the validity of our responses. The survey

methodology adheres to the guidelines of survey design, administration and validation recommended by the NOAA

Blue Ribbon Panel on CV. Validation tests were preformed on the data consist of checking for consistency

between respondents’ choices, their demographic and attitudinal characteristics, and their revealed preference

behavior.

III. Survey Instrument and Research Design

Survey Instrument

A three part survey instrument was designed to determine willingness to pay for malaria prevention. The

first part asked questions about a household’s current health status, knowledge of malaria, and expenditures on

malaria prevention and treatment. The second part presented the respondent with one of two contingent valuation

scenarios. The third part requested information on the socioeconomic characteristics of the household, including

age of respondent, education of respondent and their spouse, and household income and assets.

The contingent valuation scenarios were designed to determine households’ WTP for two types of malaria

prevention: (i) a hypothetical malaria vaccine, which would prevent each recipient from contracting malaria for

one year; and (ii) a double-size polyester bednet, which would reduce the probability of contracting malaria if

properly used. As stated earlier, both scenarios were designed according to the NOAA Panel guidelines for CV

and included a detailed description of the commodity, checked respondent understanding of how the commodity

works, and provided reminders of substitute goods and the budget constraint.

Enumerators read the contingent valuation scenario to either the head of household or their spouse and

then asked them a referendum question at one of five randomly assigned prices.4 ‘Yes’ responses were followed by

a series of questions about intrahousehold allocation of the hypothetical vaccine or bednet (Figure 1). For

example, a head of household receiving the hypothetical vaccine scenario was asked the following question:

“Suppose the price of each hypothetical vaccine against malaria was 40 Birr. Would your household buy one or

more vaccines at this price?” If they answered ‘yes’ to this question, then they were asked: “How many

vaccinations would your household buy?” This was followed by questions about how many of these vaccines

would be for adults (≥ 18 years old), for teenagers (12-17 years old), and for children (≤ 11 years old) in their

household.

Research Design

The sampling strategy employed a three stage design. In the first stage, two adjacent districts were

identified. In the second stage nine villages were chosen in each district based on reported incidence of malaria and

elevation. We used elevation as a design point because it is a good exogenous measure of malaria incidence, with

villages at higher elevations having less malaria. In the third stage, a sample of households was chosen in each

village. The latter was difficult since lists of households did not always exist or were out of date and households

were often dispersed over a wide area.

4 Prices for the hypothetical vaccine were Birr {5, 20, 40, 100, 200} and for the bednet they were Birr {8, 20, 40, 60, 100}.

US$1 is equal to Birr 6.3.

The enumerators were taken to different parts of the villages and instructed to walk in a specified direction

and interview either the self reported head or their spouse in every other household. Unfortunately, we did not

specify a protocol for who should be interviewed if both the head of household and their spouse were present. This

decision was left to the enumerator. A result, which we control for in our multivariate analysis, is that female

enumerators were significantly more likely to interview women.

Each household was presented with only one scenario, and every household in a given village was

presented with the same scenario. Six villages received the bednet scenario and twelve villages received the

hypothetical vaccine scenario (Figure 2). For this analysis, only households in which the respondent was married

and their spouse was alive were examined. Also, 41 households that did not know what malaria is and 13

households that responded ‘don’t know’ to the choice question were dropped from the sample. The final sample

size was 655 households (3,556 individuals), of which 192 (978) were offered the bednet scenario and 463 (2,578)

were offered the hypothetical vaccine scenario.

IV. Study Sample

The primary activity in our study area is subsistence cultivation. Malaria is endemic with peak

transmission taking place at the end of the rainy season. The combination of rainfed agriculture and unreliable

rainfall leads to erratic crop production; famine causing droughts have occured approximately every tenth year

throughout this century. Given the relatively small size of their holdings, the low productivity of their land, and

difficulties in gaining access to inputs and technology, many households are unable to produce enough food. This

situation is compounded by the presence of malaria, which coincides with the begining of the harvest season.

Knowledge of Malaria

Overall, men and women appear to have equal knowledge of malaria, with the majority of respondents

correctly identifying the most common ways in which the disease is transmitted. Some exceptions are that men are

more likely than women to know that mosquitoes transmit malaria; and women are more likely than men to

identify “wet places” as linked to the disease (Table 1). In terms of symptoms, both men and women correctly

identified fever and shivering as two of the most important symptoms of malaria (Table 2). Men are more likely

than women to identify joint pain as a symptom of the disease.

There are gender based differences in the respondents’ perception of whether malaria is more serious for

adults or for children (Table 3). Forty-nine percent of married men believe that malaria is more serious for

children, while 47 percent believe that it is equally serious for adults and children. Thirty-eight percent of married

women believe that malaria is more serious for children and 57 percent believe that it is equally serious for both

adults and children. Both men and women believe that taking chloroquine is the best way to cure malaria. (Table

4). There are no gender based differences in perceptions of how to avoid malaria (Table 5). Twenty-three percent

of respondents identified taking chloroquine and 79 percent identified draining wet areas as the best ways to avoid

getting malaria.

Malaria Incidence

Fifty-nine percent of our respondents claim to have had the disease at least once in the last two years

(Table 6). The average number of episodes is 2 (in the last two years) and the average number of work days

missed during the most recent episode is 24.5 In addition to themselves, 56 percent of respondents report an

5 This is quite high for an area characterized as having endemic malaria. The mean is skewed by a small number of

implausibly high responses to this question. It may be argued that the median, which is 14 for all three demographiccategories, is more representative of our sample than the mean.

average of one other adult in the household having had the disease in the last two years and 52 percent report an

average of 1 teenager or child having had the disease in the last two years.

There is an obvious disparity between the sexes in the reported incidence of malaria (65 percent for

husbands and 53 percent for wives). A number of factors might contribute to this disparity. First, men are

generally more mobile than women and children, and thus may be more exposed to infection. Second, since women

seek medical care less frequently than men, they are less likely to have confirmation of malaria diagnoses. This is

due in part to the fact that 98 percent of the community health workers are men carrying out their duties from their

homes (Ghebreyesus et al. 1996). Focus group discussions have revealed that women are reluctant to see male

community health workers for cultural reasons and thus treatment may under-report the incidence of malaria

among women (Ghebreyesus et al. 1996). Finally, the health of males may be given priority since they are required

to perform strenuous physical activities, such as plowing, to support the household’s agricultural production.

Cost of Illness

The average direct cost of treating malaria by household demographic category is shown in Table 7.

These costs represent household expenditures on medicine, health practitioners, and transportation.6 They range

from 11 Birr per episode for adults to about 5 Birr per episode for children. The largest component of these costs

is on health practitioners and the smallest on transportation. There are no significant differences in reported direct

costs between male and female respondents for themselves or for members of their households.

Socioeconomic Characteristics

Socioeconomic characteristics for the sample population are summarized in Table 8. There are significant

gender based differences in reported household agricultural income, with husbands reporting larger figures than

wives. This suggests either a systematic difference in the income of households in which men and women were

interviewed (non-equivalence of test groups) or a gender based recall error (measurement error).

Examination of non-agricultural measures of wealth such as off-farm income, housing characteristics, and

household assets (lanterns, beds, radios, shoes, and jerricans) reveals that there are no significant differences

between male and female respondents, thus discounting the possibility of non-equivalence. In Northern Ethiopia,

while both sexes are equally involved in agricultural production, men are responsible for marketing agricultural

surplus (livestock and grain) and women for minding the granary. This would explain why we observe men

reporting higher agricultural income than women: they have better recall of agricultural production while women

recall consumption.

6 This includes time spent travelling valued at 1/3 the hourly wage of unskilled labour.

Average household size for both men and women is five members, ordinarily consisting of two adults, one

teenager, and two children. That household size and composition are the same between male and female

respondents reinforces the notion that the two test groups are equivalent. The mean age of respondents is 40, with

husbands being above the mean and wives being below it. This difference is not surprising given that there has

been a civil war in the region for the last 40 years. There also appears to be a considerable disparity in literacy by

gender. Twenty-two percent of the husbands in our sample responded that they could read a newspaper easily,

while only 8 percent of wives responded in this manner.

V. Econometric Model

We now turn to the question of whether married men and married women have the same stated preferences

with regard to purchasing malaria prevention. If the answer to this question is that they are the same, then their

responses may be pooled in a unitary model. If the answer is that they are not the same, then their responses may

not be pooled.

In order to answer this question we develop a statistical model that allows us to test the pooling

assumption. To make the model operational we make the very strong assumption that the choice of whether or not

to allocate the vaccine or bednet to one household member is independent of the decision to allocate it to other

household members. Thus each family enters the statistical model n times, once for each family member.7 The

statistical model takes the following form:

Pr{yesij}= Xijββ + Zjγγ + uj+εij (1)

We estimate the model using probit, which relates the choice probability Pr{yesij} for the ith member of the jth

household to explanatory factors in such a way that the predicted probability remains in the [0,1] interval. Xij

includes dummy variables for the demographic category of the individual (e.g. adult, teenager, or child) and Zj

includes variables that vary by household and village such as price, size, income, and altitude. We correct for

within-family “clustering” effects (uj) using Stata’s cluster option. The explanatory variables used in the statistical

model are outlined in Table 10.

In order to test the hypothesis that there are no behavioral differences between men and women with

respect to the choice decision, we interact all of the variables in the model with gender of the respondent (with

female equal to one). This is equivalent to specifying separate structural equations to explain the behavior of

married men and women. The statistical model that incorporates this flexibility takes the following form:

Pr{yesij}= Xijββ + XijDjαα+ Zjγγ + ZjDjθθ+uj+εij (2)

We turn again to the question of whether behavioral characteristics of married men and women are the

same with regard to purchasing malaria prevention. Statistically, the null hypothesis is that each and every one of

variables in αα and θθ is equal to 0 and the alternative is that at least one of the variables in αα and θθ is not equal to

0. If we fail to reject the null hypothesis, then the stated preferences of men and women may be “pooled” and

7 There was a small number of enumerator and coding errors in the responses to questions about household size and

composition. These errors were resolved by adjusting the data in three ways: (a) household size is always equal to thesum of adults, teenagers, and children; (b) the number of other adults with malaria is never greater than the numberof adults in the household minus one (for the respondent); and (c) the number of teenagers and children with malariais never greater than the total number of teenagers plus children in the household.

treated as identical within the context of the statistical model. In order to test this hypothesis we estimate models

(1) and (2) and conduct a likelihood-ratio test to see if model (1) is nested in model (2).

VI. Results

Validity of Results

The principle choice question in the survey asked respondents if they would purchase one or more bednets

or hypothetical vaccines at a specified price. Respondents were randomly assigned one of five different prices.

Univariate household level responses to these questions are presented in Figure 3 and Figure 4. For both scenarios

respondents reacted to price as expected, with the percent of respondents saying “yes” to one or more bednets or

vaccines decreasing with an increase in price. The number of bednets and hypothetical vaccines the respondent

agreed to purchase declined with an increase in price. We also checked the validity of our results using

multivariate analysis and found that the coefficients in all the models have the expected sign. That respondents

reacted to the choice questions as predicted by economic theory increases our confidence in the internal and

theoretical validity of the contingent valuation scenarios.

A second check of the validity of results involves comparing revealed and stated preference results. Our

data show that, on average, out-of-pocket expenses to treat a case of malaria are between 5 and 11 Birr (Table 7).

Figure 3 and 4 indicate that 50 percent of respondents are willing to pay approximately 60 Birr for one double size

bednet. Assuming two people use a bednet for four years, then annual WTP to reduce the probability of

contracting malaria and other diseases is about 7.5 Birr per person per year. Similarly, for the hypothetical

vaccine scenario, 50 percent of respondents are willing to pay approximately 20 Birr per vaccine. The fact that

our revealed and stated preference results are concordant increases our confidence in the external validity of the

CV scenarios.

Stated Preferences

A summary of mens’ and womens’ responses to the CV questions are presented in Table 11. Men said

“yes” for 55 percent of the individuals in households receiving the bednet scenario and 26 percent in households

receiving the hypothetical vaccine scenario. Women said “yes” for 61 percent of the individuals in households

receiving the bednet scenario and 28 percent of the individuals in households receiving the hypothetical vaccine

scenario. For the bednet scenario, the difference between mens’ and womens’ responses is significant at the 5

percent level. This is because men are less likely than women to say “yes” when it comes to protecting adults with

bednets. For the hypothetical vaccine the differences between mens’ and womens’ responses are not significant.

Table 12 and Table 13 summarize mens’ and womens’ responses to the different scenarios by

demographic category and price. These are slightly different than the results presented in the previous section

because they are responses for individual rather than households. Our data suggest that, for all demogarphic

categories, 50 percent of men responded “yes” at a price of about 70 Birr per bednet and 50 percent of women

respondend “yes” at a price of 45 Birr per bednet. In the case of the hypothetical vaccine, 50 percent of men

responded “yes” at a price of approximately 10 Birr and women at a price closer to 5 Birr. This result is more

convincing than that for the bednet because of the much larger sample size for the hypothetical vaccine scenario.

More importantly, these simple univariate statistics run counter the conventional wisdom that increasing a

mother’s control of family resources (over a father’s) will result in a larger improvement in child health.

The econometric analysis for the bednet scenario is presented in Table 14 and for the hypothetical vaccine

scenario in Table 15. Four models are presented in each Table: (i) a full model, (ii) a restricted model, (iii) a model

for men only, and (iv) a model for women only. All the models appear to fit the data well. They all have χ2

statistics that are highly significant; the bednet models have pseudo R2 from 31 percent to 20 percent; and the

hypothetical vaccine models from 30 percent to 20 percent.

In the bednet scenario, the coefficients on price, income, and age are significant and have the expected

sign in all four models. Men are more likely to say “yes” to the choice question for children than for adults when

compared to women. While only significant at the 10 percent level, it would appear that ability to read, which is a

proxy for education, has a much larger positive influence on the probability saying “yes” to the choice question for

female respondents than for male respondents.

The results for the hypothetical vaccine scenario are similar to those for the bednet scenario. Price and

household size are significant in all four models. Income is significant in all models except that for women. For

this scenario we observe significant interactions between the independent variables and the gender of the

respondent. In particular, once again ability to read has a much larger positive influence on the probability saying

“yes” to the choice question for women than for men. Assuming that women can influence outcomes, this finding

is consistent with the conventional wisdom that educating women (over men) will result in improved child health.

We use a likelihood ratio test to test the null hypothesis that there are no behavioral differences between

husbands and wives with respect to the decision to purchase bednets or hypothetical vaccines. This test compares

the maximized log-likelihood of the full model to that of the restricted model. If the restriction is true, then the

maximized value of the log-likelihood of the restricted model should not be significantly less than that of the

unrestricted model.

The likelihood ratio test statistic for bednets is 36.23 and for the hypothetical vaccine 76.17. At the

α=0.05 level of significance the test critical value is χ2(12) = 21.03. Thus we conclusively reject the null for both

bednets and the hypothetical vaccine. Rejection of the null suggests that the stated preferences of husbands and

wives may not be “pooled” and therefore they may not be treated as identical in the context of the unitary model.

VII. Conclusions

Using contingent valuation questions, we reject the null hypothesis that married men and women in

Northern Ethiopia have the same stated preferences for the allocation of preventive health care in their households.

This result provides further support for shifting the “burden of proof” from the collective to the unitary models of

household behavior (Alderman et al 1995). It does not, however, tell us how child and teenager health outcomes

will differ if parents of one sex are targeted over parents of the other sex by a malaria prevention program.

Our research is unique in that it is both a new approach to testing the maintained assumptions of the

unitary model and it is a new application of the contingent valuation method. We present a new test of the

underlying assumptions of the unitary model, in which the choice questions presented to respondents are

completely exogenous to past and present household behavior.

Our methodology also opens the door for a promising new approach to testing and developing household

behavioral models. Consider, for example, Chiappori’s work on the efficient household model. This model

assumes that cooperative household decisions occur in such a way that no one individual in the household can be

made better-off without another individual being made worse-off. If it can be demonstrated that husbands and

wives are willing to pay different amounts for preventive health care and that actual purchase outcomes are

reached somewhere below the higher willingness to pay, then this is a violation of the efficient household model.

Another area in which our research is very promising is in comparing stated and revealed preferences.

Unfortunately we are unable to do this with our data because of a difference in the good observed in our stated and

revealed questions. We asked contingent valuation questions about prevention and observed household decisions

about treatment. Nonetheless, if stated preferences are thought of as the optimal outcome for the individual and

revealed preferences as the actual outcome of the household decision making process, it would be possible to

combine these two types of data to test alternative household behavioral models.

References

Alderman, Harold, P.A. Chiappori, L. Haddad, J. Hoddinott, and R. Kanbur. 1995. “Unitary VersusCollective Models of The Household: Is it Time to Shift the Burden of Proof?” The World Bank ResearchObserver, vol. 10, no. 1, pp.1-19

Behrman, J. 1997. “Intrahousehold Distribution and the Family.” Handbook of Population and FamilyEconomics. Edited by M. Rosenzweig and O. Stark. Elsevier Science B.V.

Carson, R., N. Flores, K. Martin, and J. Wright. 1997. Contingent Valuation and Revealed PreferenceMethodologies: Comparing the Estimates for Quasi-Public Goods. Land- Economics;72(1), February 1996, pp.80-99.

Chiappori, P. 1988. Rational Household Labor Supply. Econometrica. 56:1, pp. 63-89.

Chiappori, P. 1992. Collective Labor Supply and Welfare. Journal of Political Economy. 100:3, pp. 437-67.

Ghebreyesus, T., T. Alemayehu, A. Bosman, K. Witten, and A. Teklehaimanot. 1996. Communityparticipation in malaria control in Tigray region Ethiopia. Acta Tropica. 61: pp. 145-156.

Haddad, Lawrence, J. Hoddinot, and H. Alderman. 1997. Intrahousehold Resource Allocation in DevelopingCountries: Models, Methods, and Policy. L. Haddad, J. Hoddinot, and H. Alderman, eds. Johns Hopkins.Washington, D.C.

Haddad, Lawrence and Ravi Kanbur. 1990. “How Serious is Neglect of Intra-Household Inequality?” TheEconomic Journal, 100:866-881.

Hanemann, W. 1994. Welfare evaluations in contingent valuation experiments with discrete responses.American Journal of Agricultural Economics. 66, 322-341.

Hoddinott, J. and L. Haddad. 1995. Does Female Income Share Influence Household Expenditures?Evidence from the Cote D’Ivoire. Oxford Bulletin of Economics and Statistics. 57:1, 77-96.

Kanbur, R. and L. Haddad. 1994. Are Better Off Households More Unequal or Less Unequal: A BargainingTheoretic Approach to ‘Kuznets Effects’ at the Micro level. Oxford Economic Papers, July.

Lundberg, S. and R. Pollak. 1993. Separate Spheres Bargaining and the Marriage Market. Journal ofPolitical Economy. 101:6, pp. 988-1010.

Lundberg, S. and R. Pollak. 1994. Non-Cooperative Bargaining Models of Marriage. American EconomicReview Papers and Proceedings. 84:2, pp. 132-37.

Lundberg, S. and R. Pollak. 1996. Bargaining and Distribution in Marriage. Journal of Economic Perspectives.10:4, pp. 138-158.

Lundberg, S., Pollak, R. and T. Wales. 1997. Do Husbands and Wives Pool Their Resources? Evidencefrom the U.K. Child Benefit. Journal of Human Resources.

Manser, M. and M. Brown. 1980. Marriage and Household Decision Making: a Bargaining Analysis.International Economic Review. 21 :pp. 31-44.

McElroy, M. and M. Horney. 1981. Nash-Bargained Household Decisions: Towards a Generalization of theTheory of Demand. International Economic Review. 22:333-349.

Phipps, S. and P. Burton. 1992. What’s Mine is Yours? The Influence of Male and Female Incomes onPatterns of Household Expenditure. Working Paper 92-12, Department of Economics, DalhouiseUniversity.

Shultz, T. 1990. Testing the Neoclassical Model of Family Labor Supply and Fertility. Journal of HumanResources. 25(4): 599-634.

Smith, V.K. 1993. Nonmarket Valuation of Environmental Resources: An Interpretive Appraisal. LandEconomics. 69(1):1-26.

Solow, R. and K. Arrow. 1993. Report of the National Oceanic and Atmospheric Administration Panel onContingent Valuation. Federal Register, 58 (10): 4601- 4614.

Strauss, J. and D. Thomas. 1995. Human Resources: Empirical Modeling of Household Family Decisions. InHandbook of Development Economics Volume 3, eds J. Behrman and T.N Srinisvasan. Amsterdam: NorthHolland Press.

Strauss, J. and K. Beegle. 1996. Intrahousehold Allocations: A Review of Theories, Empirical Evidence andPolicy Issues. Michigan State University International Development Working Paper No. 62. Michigan StateUniversity.

Thomas, D. 1990. Intra-household Resource Allocation: An Inferential Approach. Journal of HumanResources. 25(4):635-664.

Thomas, D. 1993. Incomes, Expenditures, and Health Outcomes: Evidence on Intra-household ResourceAllocation. Washington: IFPRI mimeo.

Figure 1. Structure of Contingent Valuation Referendum Question

How many of these would be for children?

How many of these would be for teenagers?

How many of these would be for adults?

YesIf yes, how many would you buy?

No Don't Know

Would you buy one or more bednets/hypotheticalvaccines at a price of X Birr per unit?

Figure 2. Research Design

price 8n = 60

price 20n = 60

price 40n = 60

price 60n = 60

price 100n = 100

Bednet6 villagesn = 300

price 5n = 120

price 20n = 120

price 40n = 120

price 100n = 120

price 200n = 120

Hypothetical vaccine12 villages

n = 600

Tembien Sectorn = 900

Figure 3: Bednet Scenario

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

8 20 40 60 100

prices

% re

spon

dent

s

No

Yes

Figure 4: Hypothetical Vaccine Scenario

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

5 20 40 100 200

price

% re

spon

dent

s

No

Yes

Table 1: Malaria transmissionn All Men Women

Mosquitos 655 72% 77% 67%*Lack of sanitation 655 29% 27% 30%Impure water 655 19% 19% 19%Wet places 655 45% 42% 50%*Other 655 4% 4% 5%Don't know/not sure 655 5% 4% 7%* indicates difference between men and women is significant at the 5 percent level

Table 2: Symptoms of malarian All Men Women

Fever 655 73% 75% 72%Shivering 655 92% 92% 92%Convulsions 655 7% 8% 5%Headache 655 33% 35% 31%Backache 655 30% 29% 30%Anemia 655 2% 1% 3%Stomachache 655 4% 4% 4%Vomiting 655 46% 46% 47%Diarrhea 655 6% 6% 5%Joint pain 655 31% 37% 25%*Loss of appetite 655 17% 18% 16%Other 655 12% 14% 10%* indicates difference between men and women is significant at the 5 percent level

Table 3: Seriousness of malaria for adults and childrenn All Men Women

Adults 654 4% 4% 4%Children 654 44% 49% 38%*Serious for adults and children 654 52% 47% 57%*Not serious for adults and children 654 0% 0% 0%* indicates difference between men and women is significant at the 5 percent level

Table 4: Best ways to cure malarian All Men Women

Taking chloroqine 655 59% 61% 57%Taking other medicine 655 3% 3% 3%Injections 655 53% 53% 53%Changing one's diet 655 5% 5% 5%Cleaning 655 5% 4% 6%Religious healing 655 4% 3% 4%herbal medicines 655 3% 4% 2%* indicates difference between men and women is significant at the 5 percent level

Table 5: Best ways to avoid getting malarian All Men Women

There is no way to prevent malaria 655 1% 2% 1%Cleaning house / wet areas 655 79% 79% 78%Sleeping under a bednet 655 1% 1% 2%Taking chloroquine 655 23% 24% 22%Eating a special diet 655 11% 10% 12%Spraying insecticides 655 11% 10% 12%Other 655 3% 3% 3%Don't know/not sure 655 6% 4% 8%* indicates difference between men and women is significant at the 5 percent level

Table 6: Malaria IncidenceVariable n All Men WomenRespondent has had malaria in last two years 655 59% 65% 53%*No. times the respondent had malaria in last two years 510 2 2 2No. days respondent missed work (most recent episode) 353 24 23 25Households in which other adults have had malaria 651 56% 53% 60%No. of other adults that have had malaria 651 1 1 1Household in which teenagers or children had malaria 654 52% 53% 51%No. of teenager and children that have had malaria 654 1 1 1* indicates difference between men and women is significant at the 5 percent level

Table 7: Average Direct Cost of Illness for Most Recent Episode of Malaria (in Birr)Male Respondent Female Respondent

Adult 12.22 9.85Teenager Female Male

5.998.963.89

6.769.575.76

Child Female Male

4.885.016.67

6.325.884.81

* indicates difference between men and women is significant at the 5 percent level

Table 8: Household income and assetsVariable n All Men WomenAnnual agricultural income 654 1175 1259 1085*Annual wage income 471 326 320 333Number of rooms 654 1.2 1.3 1.2Number of buildings 651 2.4 2.5 2.4lamp 655 1.1 1.1 1.1bed 655 0.14 0.13 0.14radio 655 0.26 0.34 0.18bicycle 654 0.02 0.02 0.01shoes 323 0.98 1.01 0.94jerrican 322 0.5 0.5 0.5* indicates difference between men and women is significant at the 5 percent level

Table 9: Household characteristicsVariable n All Men WomenHousehold size 655 5 5 5Number of adults 655 2 2 2Number of teenagers 655 1 1 1Number of children 655 2 2 2Age of respondent 655 40 46 34*Respondent can read a newspaper easily 655 15% 22% 8%*Respondent’s spouse can read a newspaper easily 653 17% 3% 32%** indicates difference between men and women is significant at the 5 percent level

Table 10: Description of variables used in multivariate analysisVariable Description Unit

YESNO Decision maker’s response to contingent valuation scenario forindividual i in household j.

yes = 1no = 0

CHILD Whether individual i in household j is a child. child = 1

TEEN Whether individual i in household j is a teenager. teen = 0

LNPRICE Log of price presented to the respondent in the contingentvaluation scenario.

Birr

HHSIZE Number of individuals in household j individuals

HHCOI Expected household direct cost of illness8 Birr

LNINC Log annual household income (agricultural income plus off-farm wage income).

Birr

NOWAGE Missing observation on wage income. wage=0no wage=1

SREAD Whether the respondent’s spouse can read a newspaper easily. read=1no read=0

RREAD Whether the respondent can read a newspaper easily. read=1no read=0

AGE Age of the respondent. years

ALT Altitude of village meters

NAME1 sex of enumerator 0 = male1 = female

GENDER Dichotomous variable for respondent gender 0 = male1 = female

8 Calculated in the following manner: total number of adults that have had malaria at least once in the last two years times

11 Birr plus the number of children and teenagers that have had malaria in the last two years times 6 Birr.

Table 11: Percentage ‘Yes’ responses by demographic category.Bednets Hypothetical Vaccine

Men Women Men Women

Adults 49% 58%** 27% 27%

Teenagers 49% 60% 23% 30%*

Children 63% 65% 26% 28%

Total 55% 61%** 26% 28%

* indicates difference significant at the 10 percent level and ** at the 5 percent level

Table 12: Percentage “Yes” responses for Bednets by price and demographic category

Bednets Men Women

Price Adults Teenagers Children Adults Teenagers Children

8 75% 100% 97% 78% 87% 75%

20 56% 50% 77% 83% 100% 92%

40 58% 60% 63% 52% 44% 55%

60 60% 67% 62% 31% 44% 44%

100 21% 19% 29% 58% 44% 61%

Table 13: Percentage “yes” responses for vaccines by price and demographic categotyVaccines Men Women

Price Adults Teenagers Children Adults Teenagers Children

5 70% 63% 70% 50% 53% 49%

20 41% 33% 36% 33% 39% 38%

40 16% 13% 15% 22% 20% 21%

100 15% 9% 13% 18% 17% 15%

200 3% 2% 6% 8% 0% 8%

TABLE 14: BEDNET SCENARIO - FULL MODELProbit Estimates Number of obs = 935 chi2(25) = 84.64 Prob > chi2 = 0.0000Log Likelihood = -473.15621 Pseudo R2 = 0.2531

(standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .5222958 .1369528 3.814 0.000 .2538732 .7907184 teen | -.0004991 .2575952 -0.002 0.998 -.5053763 .5043782 lnp | -.8473361 .1912231 -4.431 0.000 -1.222127 -.4725457 hhsize | -.079234 .1108 -0.715 0.475 -.2963979 .13793 hhcoi | .0107989 .0114936 0.940 0.347 -.0117282 .0333259 lninc | .762905 .3015215 2.530 0.011 .1719338 1.353876 nowage | .459336 .3421437 1.343 0.179 -.2112534 1.129925 rread | .1830093 .352781 0.519 0.604 -.5084287 .8744473 sread | -.61187 .7487201 -0.817 0.414 -2.079334 .8555944 age | -.0469432 .0133319 -3.521 0.000 -.0730732 -.0208132 alt | .0012243 .0008922 1.372 0.170 -.0005243 .0029729 name1 | .2828283 .3442089 0.822 0.411 -.3918088 .9574654 gxchild | -.4183886 .1790632 -2.337 0.019 -.7693459 -.0674312 gxteen | .2150069 .2974898 0.723 0.470 -.3680624 .7980762 gxlnp | .2654721 .2446805 1.085 0.278 -.2140929 .7450372gxhhsize | -.004067 .1423619 -0.029 0.977 -.2830913 .2749572 gxhhcoi | -.0022124 .0153785 -0.144 0.886 -.0323538 .027929 gxlninc | -.5485082 .3171713 -1.729 0.084 -1.170153 .0731362gxnowage | -.2240044 .4376035 -0.512 0.609 -1.081691 .6336827 gxrread | .4389155 .4970089 0.883 0.377 -.535204 1.413035 gxsread | .7614102 .8167001 0.932 0.351 -.8392925 2.362113 gxage | .0111303 .0194497 0.572 0.567 -.0269904 .0492511 gxalt | -.000675 .0013357 -0.505 0.613 -.0032929 .0019428 gxname1 | -.2074089 .4782649 -0.434 0.665 -1.144791 .7299732 gender | 3.395596 3.177593 1.069 0.285 -2.832372 9.623564 _cons | -2.235173 2.532551 -0.883 0.377 -7.198882 2.728537------------------------------------------------------------------------------

TABLE 14: BEDNET SCENARIO - RESTRICTED MODELProbit Estimates Number of obs = 935 chi2(12) = 59.00 Prob > chi2 = 0.0000Log Likelihood = -491.27213 Pseudo R2 = 0.2245

(standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .2934542 .0913876 3.211 0.001 .1143378 .4725706 teen | .1297512 .1373331 0.945 0.345 -.1394167 .3989191 lnp | -.6303649 .1199329 -5.256 0.000 -.8654292 -.3953007 hhsize | -.0669207 .0679727 -0.985 0.325 -.2001446 .0663033 hhcoi | .0090063 .007675 1.173 0.241 -.0060363 .024049 lninc | .2873795 .0943526 3.046 0.002 .1024518 .4723072 nowage | .2345994 .2193079 1.070 0.285 -.1952361 .6644349 rread | .5316463 .2421548 2.195 0.028 .0570316 1.006261 sread | .0504008 .2839599 0.177 0.859 -.5061504 .6069519 age | -.0312152 .0090719 -3.441 0.001 -.0489958 -.0134346 alt | .0006408 .0006339 1.011 0.312 -.0006016 .0018832 name1 | .1012184 .2365512 0.428 0.669 -.3624133 .5648502 _cons | .4750495 1.228571 0.387 0.699 -1.932905 2.883004------------------------------------------------------------------------------

TABLE 14: BEDNET SCENARIO - MARRIED MEN ONLYProbit Estimates Number of obs = 412 chi2(12) = 53.05 Prob > chi2 = 0.0000Log Likelihood = -194.50266 Pseudo R2 = 0.3133

(standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .5222958 .1374247 3.801 0.000 .2529483 .7916432 teen | -.0004991 .2584827 -0.002 0.998 -.5071159 .5061177 lnp | -.8473361 .191882 -4.416 0.000 -1.223418 -.4712543 hhsize | -.079234 .1111817 -0.713 0.476 -.2971462 .1386782 hhcoi | .0107989 .0115332 0.936 0.349 -.0118058 .0334035 lninc | .762905 .3025604 2.521 0.012 .1698975 1.355912 nowage | .459336 .3433226 1.338 0.181 -.213564 1.132236 rread | .1830093 .3539965 0.517 0.605 -.510811 .8768296 sread | -.61187 .7512998 -0.814 0.415 -2.084391 .8606506 age | -.0469432 .0133778 -3.509 0.000 -.0731632 -.0207231 alt | .0012243 .0008952 1.368 0.171 -.0005303 .0029789 name1 | .2828283 .3453949 0.819 0.413 -.3941333 .9597899 _cons | -2.235173 2.541277 -0.880 0.379 -7.215984 2.745639------------------------------------------------------------------------------

TABLE 14: BEDNET SCENARIO - MARRIED WOMEN ONLYProbit Estimates Number of obs = 523 chi2(12) = 30.15 Prob > chi2 = 0.0027Log Likelihood = -278.65355 Pseudo R2 = 0.2002

(standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .1039072 .115605 0.899 0.369 -.1226744 .3304888 teen | .2145078 .1491298 1.438 0.150 -.0777812 .5067968 lnp | -.581864 .1529772 -3.804 0.000 -.8816937 -.2820343 hhsize | -.083301 .0895795 -0.930 0.352 -.2588735 .0922715 hhcoi | .0085865 .0102393 0.839 0.402 -.0114822 .0286551 lninc | .2143968 .0986098 2.174 0.030 .0211252 .4076684 nowage | .2353316 .2734101 0.861 0.389 -.3005423 .7712055 rread | .6219248 .350839 1.773 0.076 -.065707 1.309557 sread | .1495402 .3269139 0.457 0.647 -.4911994 .7902797 age | -.0358128 .0141919 -2.523 0.012 -.0636285 -.0079971 alt | .0005493 .0009961 0.551 0.581 -.0014031 .0025016 name1 | .0754194 .3327604 0.227 0.821 -.5767789 .7276178 _cons | 1.160423 1.92329 0.603 0.546 -2.609157 4.930003------------------------------------------------------------------------------

TABLE 15: HYPOTHETICAL VACCINE SCENARIO - FULL MODELProbit Estimates Number of obs = 2511 chi2(24) = 155.74 Prob > chi2 = 0.0000Log Likelihood = -1083.5482 Pseudo R2 = 0.2592

(standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .0539816 .0770473 0.701 0.484 -.0970283 .2049915 teen | .0164911 .1009931 0.163 0.870 -.1814518 .214434 lnp | -.626231 .0844309 -7.417 0.000 -.7917125 -.4607495 hhsize | -.2369907 .0629908 -3.762 0.000 -.3604503 -.113531 hhcoi | -.0028938 .005823 -0.497 0.619 -.0143067 .008519 lninc | .3681772 .1652187 2.228 0.026 .0443546 .6919999 nowage | -.2290877 .2219336 -1.032 0.302 -.6640695 .2058942 rread | -.0180757 .2264677 -0.080 0.936 -.4619442 .4257928 sread | -.2161176 .7145692 -0.302 0.762 -1.616648 1.184412 age | -.0061447 .0063652 -0.965 0.334 -.0186203 .0063309 alt | -.0004162 .0006452 -0.645 0.519 -.0016808 .0008484 gxchild | .084358 .1125746 0.749 0.454 -.1362841 .305 gxteen | .1790064 .1530371 1.170 0.242 -.1209408 .4789536 gxlnp | .2203496 .1152073 1.913 0.056 -.0054527 .4461518gxhhsize | .0163739 .0858517 0.191 0.849 -.1518923 .1846401 gxhhcoi | .0241046 .0079742 3.023 0.003 .0084756 .0397337 gxlninc | -.2192301 .2077138 -1.055 0.291 -.6263416 .1878813gxnowage | .1516615 .3242403 0.468 0.640 -.4838379 .7871609 gxrread | .8758159 .42356 2.068 0.039 .0456536 1.705978 gxsread | .4811238 .7478513 0.643 0.520 -.9846378 1.946885 gxage | -.0024529 .011454 -0.214 0.830 -.0249024 .0199966 gxalt | -.0007992 .0009674 -0.826 0.409 -.0026954 .0010969 gender | 1.407346 2.062665 0.682 0.495 -2.635403 5.450095 name1 | .4172823 .191587 2.178 0.029 .0417786 .792786 _cons | 1.273433 1.312252 0.970 0.332 -1.298534 3.845399------------------------------------------------------------------------------

TABLE 15: HYPOTHETICAL VACCINE SCENARIO - RESTRICTED MODELProbit Estimates Number of obs = 2511 chi2(12) = 115.73 Prob > chi2 = 0.0000Log Likelihood = -1121.6328 Pseudo R2 = 0.2331

(standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .0808225 .0562627 1.437 0.151 -.0294503 .1910954 teen | .1031157 .0732437 1.408 0.159 -.0404392 .2466707 lnp | -.5198782 .0577314 -9.005 0.000 -.6330297 -.4067267 hhsize | -.221705 .0447392 -4.956 0.000 -.3093922 -.1340178 hhcoi | .0097027 .0041343 2.347 0.019 .0015996 .0178058 lninc | .2297162 .1073339 2.140 0.032 .0193455 .4400868 nowage | -.1072544 .1636158 -0.656 0.512 -.4279355 .2134267 rread | .2484432 .1807448 1.375 0.169 -.1058101 .6026964 sread | .2030635 .1946915 1.043 0.297 -.1785249 .5846518 age | -.0067883 .0048084 -1.412 0.158 -.0162125 .002636 alt | -.0010359 .0004729 -2.190 0.028 -.0019629 -.000109 name1 | .4097896 .1949091 2.102 0.036 .0277747 .7918044 _cons | 2.507805 1.014903 2.471 0.013 .5186302 4.496979------------------------------------------------------------------------------

TABLE 15: HYPOTHETICAL VACCINE SCENARIO - MARRIED MEN ONLYProbit Estimates Number of obs = 1385 chi2(12) = 87.21 Prob > chi2 = 0.0000Log Likelihood = -556.18634 Pseudo R2 = 0.3046 (standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .0540498 .0772674 0.700 0.484 -.0973916 .2054912 teen | .0165912 .1011569 0.164 0.870 -.1816726 .2148551 lnp | -.627405 .0854335 -7.344 0.000 -.7948516 -.4599585 hhsize | -.2378827 .0636524 -3.737 0.000 -.3626392 -.1131262 hhcoi | -.0028556 .0058634 -0.487 0.626 -.0143476 .0086363 lninc | .3745039 .1704802 2.197 0.028 .0403688 .7086391 nowage | -.2349756 .2270615 -1.035 0.301 -.680008 .2100568 rread | -.0199686 .2256689 -0.088 0.929 -.4622715 .4223343 sread | -.2170974 .7159102 -0.303 0.762 -1.620256 1.186061 age | -.0061323 .0063792 -0.961 0.336 -.0186352 .0063707 alt | -.0004057 .0006529 -0.621 0.534 -.0016855 .000874 name1 | .4598039 .2904465 1.583 0.113 -.1094609 1.029069 _cons | 1.213468 1.374699 0.883 0.377 -1.480892 3.907829------------------------------------------------------------------------------

TABLE 15: HYPOTHETICAL VACCINE SCENARIO - MARRIED WOMEN ONLYProbit Estimates Number of obs = 1126 chi2(12) = 66.52 Prob > chi2 = 0.0000Log Likelihood = -527.26925 Pseudo R2 = 0.2043 (standard errors adjusted for clustering on quesnum)------------------------------------------------------------------------------ | Robust yesno1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]---------+-------------------------------------------------------------------- child | .1374253 .0823702 1.668 0.095 -.0240173 .2988679 teen | .1956263 .1153402 1.696 0.090 -.0304363 .421689 lnp | -.4053098 .0794241 -5.103 0.000 -.5609782 -.2496414 hhsize | -.2190462 .0610364 -3.589 0.000 -.3386754 -.099417 hhcoi | .0210589 .0055751 3.777 0.000 .010132 .0319858 lninc | .1436617 .1357977 1.058 0.290 -.1224969 .4098204 nowage | -.0734722 .2462846 -0.298 0.765 -.5561811 .4092367 rread | .862611 .3590284 2.403 0.016 .1589282 1.566294 sread | .262959 .2197523 1.197 0.231 -.1677475 .6936655 age | -.0087181 .0095519 -0.913 0.361 -.0274396 .0100034 alt | -.0012374 .00073 -1.695 0.090 -.0026681 .0001933 name1 | .3817052 .2534351 1.506 0.132 -.1150185 .8784289 _cons | 2.7575 1.665288 1.656 0.098 -.5064045 6.021404------------------------------------------------------------------------------

ESPL:\JULIAN\PERSONAL\PRDP\PRWP2.DOCMay 18, 1998 9:26 AM


Recommended