Estimating reservation wages of employed workers using a stochastic frontier.

Murphy, Kevin J.

I. Introduction

According to search theory, workers form "reservation" wages such that employment offers paying wages less than the reservation wage are rejected. The first job offer paying a wage higher than the reservation wage is accepted and search is terminated. Though the reservation wage is clearly a matter of individual taste and unobservable, a number of empirical articles have nevertheless appeared on the subject. The present paper adds to this literature. Its novel aspects are that it uses data on employed workers rather than unemployed workers (as is usually the case) and that it applies a comparatively new econometric technique, stochastic frontier estimation, to the problem of estimating reservation wages.

Our interests in this paper are twofold. First, the paper provides estimates of worker reservation wages utilizing the stochastic frontier regression technique. Second, we use these estimates to test various hypotheses regarding the response of reservation wages to factors that search theory suggests are important in the determination of reservation wages.

The remainder of the paper is organized as follows. Section II discusses the previous empirical literature on the reservation wage. Section III sketches the theory and estimation techniques we use. Section IV lays out a framework for assessing the validity of the approach used here. Finally, section V presents the empirical results.

II. Background

Studies of reservation wages fall into one of two categories. One type of study utilizes survey responses to a question that directly asks the respondent what his or her reservation wage is. Studies in this category are: Crosslin and Stevens |4~, Lancaster and Chesher |20~, Feldstein and Poterba |5~, Lancaster |19~, Holzer |12~, Jones |14; 15~, and Heywood and White |8~. The second type of study of the reservation wage takes the reservation wage as an unobservable and attempts to infer it econometrically. Papers employing this approach are: Kiefer and Neumann |16; 17; 18~ and Fishe |6~. Both types of study, whether the reservation wage is a survey response or is inferred econometrically, use data on workers who are either experiencing a spell of unemployment or who have recently completed a spell of unemployment.

We regard the studies in which reservation wages are inferred econometrically to be somewhat superior on methodological grounds. This is because the reservation wage, regardless of what answer a respondent might give on a questionnaire, is not likely to be measured with accuracy. There are a number of reasons why reservation wages culled from labor market surveys may be biased. First, survey respondents when confronted with the question, "What is the lowest amount of take-home pay that you would be prepared to accept from a new job?" may simply not have a good idea of what the true answer to this question is. They may, for example, engage in wishful thinking and respond with an amount higher than that which would actually be necessary to entice them to accept new employment if confronted with a realistic opportunity.

A second source of response bias, interrelated with the first source, is that it is very difficult on a survey to control for other characteristics that a job might possess. Thus a worker confronted with an opportunity in his or her industry or occupation of choice might be willing to accept a lower wage than otherwise. The British studies |14; 15; 19; 20~ suffer from this problem in particular when compared to the American work |5; 8; 12~. The former utilize data in which a single question is asked regarding the reservation wage, worded to the effect, "What is the lowest amount in take-home pay you would be prepared to accept for a new job?" The American studies use data that rely on a two-tiered question regarding the reservation wage. The questions are generally worded in this fashion: (1) "What type of work have you been looking for?" and (2) "What would the wage or salary have to be for you to be willing to take employment in this type of work?" Thus the latter type of two-tiered approach at least confronts the respondent with a more clearly defined question.

A third source of bias in reservation wages recorded in survey data is that in most of the instances the questions are posed by a government agency. Feldstein and Poterba |5~, for example, use data collected in a supplement of the Current Population Survey. Crosslin and Stevens |4~ and Heywood and White |8~ use data collected by state employment security agencies. Only Holzer |12~ and Jones |14; 15~ clearly use data not collected by a government agency. The old saw in the response bias literature is that people will answer a question about income differently depending on whether they think the answer will affect their taxes or their credit. This cliche is no doubt applicable in the case at hand, as it is the government that administers unemployment insurance benefits. Receipt of unemployment insurance benefits is conditioned by the recipient demonstrating availability and active search for work. Thus a UI recipient may feel it's in his or her best interest to provide a low figure in response to the reservation wage question to convince the government of the seriousness of the intent to find work. In actual fact, however, the recipient may be quite content to receive benefits for the full duration of the eligibility period unless a really good offer were to materialize.

A fourth source of inaccuracy in data sets using worker reported reservation wages is non-response bias. Not all studies using this type of data discuss this problem, but two that do are |5~ and |8~. Feldstein and Poterba report a non-response rate of 31% in their data set which was collected as a supplement to the May 1976 CPS. Even more remarkably, Heywood and White report a non-response rate of 67% in their data set which was collected by the Wisconsin Job Service for Milwaukee County in November of 1984. Other authors using this type of data don't discuss the non-response problem, but, on the basis of those that do, the problem appears to be fairly serious. If respondents select out of the sample in a systematic way, then the repercussions for the validity of the reservation wage data may be very significant.

In light of the various possible sources of inaccuracy in self-reported reservation wage data, it is our conviction that empirical studies relying on such data may be flawed by systematic error. If a suitable econometric technique existed to deduce reservation wages from observed labor market data, more reliable inferences regarding the determinants of reservation wages could be obtained. As noted above, |16; 17; 18~, and |6~ do employ an econometric technique to deduce reservation wages. Specifically, these papers use data on completed and incomplete spells of unemployment in conjunction with Heckman's |7~ censored regression model to estimate a reservation wage equation. Our paper will employ a similar philosophical approach, i.e., estimation of the reservation wage, though we shall employ a different econometric technique to obtain the estimates. A second distinguishing feature of our research is the data set we use. All reservation wage studies to date have utilized data on unemployed or recently unemployed workers and therefore may not be representative of the labor market experiences of the overall work force. Kiefer and Neumann, in particular, note that their data set (used in all three of their papers) is "not representative of the entire U.S. population or even of the unemployed population" |17, 196~. Our paper instead will estimate reservation wages of workers who are employed. Specifically, we use data from the Current Population Survey. This data set has the advantage that it is intentionally designed to be representative of the U.S. population.(1)

III. Search Theory, the Reservation Wage, and Stochastic Frontier Estimation

The reservation wage enjoys a long and distinguished history in the search theory literature. It is that wage necessary to induce somebody to accept an offer of employment. The main problem with the concept is that it is unobservable as far as the economic analyst is concerned. What the economic analyst does observe is the actual wage paid to a given worker. This wage, by definition of the reservation wage, is necessarily equal to or greater than the reservation wage.

It has been demonstrated in a number of instances that a reservation wage approach or strategy to labor market search is an optimal strategy from the worker's viewpoint in the sense that it maximizes expected future income |22; 23; 21~. In this literature, the optimal reservation wage is determined by the worker equating the marginal benefit with the marginal cost of further increment to the reservation wage.

We actually observe a given worker's wage but we cannot observe the worker's reservation wage. The following model summarizes the wage determination process for an individual worker:

|w.sub.i~ = |X.sub.i~|Beta~ + ||Gamma~.sub.i~ + ||Delta~.sub.i~ (1)

where

|X.sub.i~ is a row vector of individual specific wage determining characteristics;

|Beta~ is a column vector of regression coefficients;

||Gamma~.sub.i~ is a random error term specific to individual i for which E(||Gamma~.sub.i~) = 0 and |Mathematical Expression Omitted~; and

||Delta~.sub.i~ is also an individual specific error term but is |greater than or equal to~ 0, E(||Delta~.sub.i~) |is greater than~ 0, and |Mathematical Expression Omitted~.

Interpretations of |X.sub.i~ and |Beta~ are straightforward and reflect respectively factors directly tied to the wage determination process and the systematic impacts that these factors have on worker wages. The interpretation of ||Gamma~.sub.i~ is also conventional and reflects unmeasurable wage determining factors such as native intelligence and sheer luck (or lack thereof).

The distinguishing aspect of equation (1) is the ||Delta~.sub.i~ term. It is non-negative and it reflects the degree by which the worker's observed wage exceeds the unobserved reservation wage. From (1) the worker's reservation wage is:

|Mathematical Expression Omitted~

As long as one can measure ||Delta~.sub.i~, estimation of the reservation wage is straightforward. The econometric problem of course is disentangling ||Delta~.sub.i~ from ||Gamma~.sub.i~. In a standard OLS wage regression for example the best one could do is estimation of:

|w.sub.i~ = |X.sub.i~|Beta~ + ||Epsilon~.sub.i~ (3)

where ||Epsilon~.sub.i~ = ||Gamma~.sub.i~ + ||Delta~.sub.i~. Because ||Delta~.sub.i~ is imbedded in the error term, it is impossible to deduce the worker's reservation wage from such an equation.

An econometric technique that is well-suited to decomposition of |Epsilon~, however, is stochastic frontier estimation. In its initial applications the technique was used to estimate production functions and to determine efficiency levels in such production processes.(2) But the technique is appropriate in many other instances where the dependent variable is constrained from above or below for one reason or another. In the case at hand, the dependent variable is constrained from below because of workers' self-imposed search rule, and we observe wages of workers of a given skill level distributed by varying degrees above their reservation wage.

Our work in this regard follows directly from several earlier articles treating specific inefficiencies in the labor market. Hofler and Polachek |9; 10~ defined "ignorance" as the difference between the price an individual would pay with full information and the price actually paid, given that individual's limited information. Substituting "employer" for "individual" and "wage" for "price" yields a definition of the spread between the actual wage paid and the reservation wage. This next step was taken in the innovative work by Polachek and Yoon |25~. They define employer ignorance as the gap between the reservation wage and the wage that the firm actually pays. They measure mean employer ignorance without measuring reservation wages, however. They do compare ignorance results by numerous strata within their sample (e.g., high school education vs. college education) but do not test the significance of differences across those strata. Our work here proceeds from Polachek and Yoon by estimating the reservation wage of each worker in the sample. We also concentrate on how the reservation wage measure, rather than ignorance, is consistent with search theory.(3) Finally, we extend the validation side of the Polachek and Yoon paper by testing for the significance of differences in reservation wages across numerous strata.

Estimation of ||Delta~.sub.i~, which is critical to calculation of a worker's reservation wage, requires two steps. Estimates of the parameters of the systematic part of the wage equation are made first. Using these estimates, the residuals from the first stage regression are taken yielding an approximation of the ||Epsilon~.sub.i~. From this point the ||Delta~.sub.i~ are generated through use of the conditional distribution of |Delta~ given |Epsilon~. Once this step is accomplished calculation of individual reservation wages is straightforward. We now discuss the specifics of these steps.

Recall that:

||Epsilon~.sub.i~ = ||Gamma~.sub.i~ + ||Delta~.sub.i~ (4)

where ||Gamma~.sub.i~ is iid |Mathematical Expression Omitted~ and ||Delta~.sub.i~ |is greater than or equal to~ 0. Because Monte Carlo studies (e.g., Aigner, Lovell, and Schmidt |1~ and Cowing, Reifschneider, and Stevenson |3~) indicate little difference in coefficient estimates generated by assuming a half-normal distribution for |Delta~ rather than by assuming other distributions such as the exponential, we assume ||Delta~.sub.i~ to be iid |Mathematical Expression Omitted~ and truncated at zero from below.

We can test this decision, however, by estimating three models which are distinguished by the distribution of each model's one-sided error term. The first model, of course, has ||Delta~.sub.i~ iid |Mathematical Expression Omitted~ and truncated at zero from below. The second model specifies ||Delta~.sub.i~ as exponential. The third choice is ||Delta~.sub.i~ iid |Mathematical Expression Omitted~ where |Mu~ |is not equal to~ 0 and the distribution is truncated at zero from below. Each of the models will be estimated by maximum likelihood. Likelihood ratio tests will be conducted to determine which of these three choices is the best model.

This study's approach is to estimate the model by maximum likelihood and then compute individual-specific ||Delta~.sub.i~ estimates. Jondrow, Lovell, Materov, and Schmidt |13~ show that an estimate of ||Delta~.sub.i~ can be expressed as the expected value of ||Delta~.sub.i~ given ||Epsilon~.sub.i~:

E(||Delta~.sub.i~|where~||Epsilon~.sub.i~) = |(||Sigma~.sub.|Delta~~||Sigma~.sub.|Gamma~~)/|Sigma~~ {|f(||Epsilon~.sub.i~|Gamma~/|Sigma~)/(1 - F(-||Epsilon~.sub.i~|Gamma~/|Sigma~))~ - (|Gamma~||Epsilon~.sub.i~)/|Sigma~}. (5)

Since |Gamma~(=||Sigma~.sub.|Delta~~/||Sigma~.sub.|Gamma~~), |Mathematical Expression Omitted~, and |Mathematical Expression Omitted~ are estimated directly and ||Epsilon~.sub.i~ can be estimated for each observation, person-specific ||Delta~.sub.i~ estimates can be obtained from (5). The reservation wage |Mathematical Expression Omitted~ can then be formed by subtracting the estimate of ||Delta~.sub.i~ from the actual wage |w.sub.i~. This gives a reservation wage for each person in the sample.

IV. Model Validation

In order to check the consistency of our model with search theory, we will estimate mean reservation wages for several groups for which theory presents clear hypotheses regarding reservation wage behavior. We will then compare those mean reservation wages for conformity with theory's predictions about differences across groups.

As noted above, a given worker sets the reservation wage such that the marginal benefit and marginal cost from further increment in |w.sup.r~ just balance. Determinants of the reservation wage, then, are simply the shift factors in the marginal benefit and marginal cost schedules. These shift factors have been well delineated in the theoretical work on search.(4) Marginal benefits of further increment to the reservation wage are greater lifetime earnings once employment is secured. The height of the marginal benefit schedule depends on factors such as the individual's discount rate and time horizon. The cost of further increment to the reservation wage consists of the higher out of pocket costs from an extended search and additional foregone earnings resulting from a longer duration of unemployment. The height of the marginal cost curve depends on the worker's search efficiency, skill level (which determines the opportunity cost), wealth level, and the availability and amount of social welfare payments (such as unemployment benefits) that can be collected during a spell of unemployment.

Because workers differ in terms of the various factors that determine the marginal benefits and marginal costs of further increment to the reservation wage, the model implies a number of testable hypotheses about reservation wages across workers. We test the following hypotheses in the empirical work reported below.

HYPOTHESIS 1. Demographic groups with high rates of time preference will set lower reservation wages than groups with low rates of time preference.

This hypothesis follows because the rate of time preference is inversely related to the marginal benefit locus in Figure 1; therefore, those workers with low rates of time preference will set higher reservation wages. It is conventional to associate high rates of time preference with demographic groups weakly attached to the labor force such as teens and females. Teens are in that stage of the life cycle where familial responsibilities are low and where they have typically not yet decided on a career path. Females may exhibit weak labor force attachment because of plans to bear children at some point in the future. Conversely, we expect to find that prime-age, married males exhibit higher reservation wages than other groups in the labor force. This is because familial responsibilities are higher than for, say, teens and because males typically don't leave the labor force to raise children.

HYPOTHESIS 2. Workers in areas paying relatively higher unemployment benefits will exhibit higher reservation wages.

This prediction follows because higher unemployment benefits reduce the marginal cost associated with a higher reservation wage.

HYPOTHESIS 3. More educated workers will set higher reservation wages.

We expect both discount rates and search costs among more educated workers to be lower. The former leads to a higher marginal benefit schedule and the latter implies a lower marginal cost schedule. Taken together, both factors induce a higher reservation wage. Discount rates should be lower for more educated workers because they have, in a sense, revealed them to be lower by spending additional years foregoing income while accumulating human capital. Search costs for more educated workers should be lower for two reasons. First, more educated workers are likely to be more efficient searchers. Second, and perhaps along similar lines, we expect that more educated workers have access to better information networks than less educated workers.

HYPOTHESIS 4. Workers in urban areas will have higher reservation wages.

Firms are less dispersed geographically and transport networks are generally better in urban areas. Search costs are lower for both reasons, implying higher reservation wages.

HYPOTHESIS 5. Workers with greater wealth will exhibit higher reservation wages.

Wealthier workers have more non-labor income. The cost of search for such workers is lower as a consequence and therefore we expect wealthier workers to set higher reservation wages. We use home ownership status as a proxy for wealth in the empirical work reported below.

V. Empirical Results

Our estimation strategy consists of the following five steps:

1. Estimate a single frontier equation on the entire sample.

2. Generate |w.sup.r~ for each person in the sample on the basis of equation (5).

3. Partition the entire sample into various cells according to the determinants of |w.sup.r~.

4. Calculate the mean |w.sup.r~ for each cell.

5. Compare the mean |w.sup.r~ across the cells using analysis of variance techniques.

We will elaborate on step 5 for a moment. A test for differences in reservation wages can be conducted by comparing mean |w.sup.r~ across the cells. Two types of test are employed in order to evaluate the statistical significance of the results. Whenever the data are partitioned into only two groups (e.g., female vs. male), t-ratios are calculated to test the null hypothesis that the mean reservation wage level of the two groups are equal.(5) Whenever the data are partitioned into more than two groups, an analysis of variance procedure is used. First the mean |w.sup.r~ for each group is computed, then an F-statistic is computed to test the equality of all of the means. Whenever this TABULAR DATA OMITTED step reveals significant differences among the means, 95% confidence intervals are constructed around the difference in means of each pair of |w.sup.r~. These are then used to decide whether or not the difference in means of each pair is statistically significant.

The data used in the empirical work reported below were drawn from the January 1983 Current Population Survey tape. The analysis was confined to individuals who were working full time during the survey week. Explanatory variables used in estimating the systematic part of the frontier equation include the usual factors found in the standard human capital wage equation, such as the worker's education, age, marital status, geographic location, gender, and so forth. The dependent variable is the natural logarithm of the worker's hourly wage. The data appendix gives a full accounting of the variables used in the analysis.

As mentioned earlier, three different models were estimated by maximum likelihood using, respectively, the half-normal, exponential, and truncated normal one-sided error terms |Delta~. The first and third modes converged whereas the second did not. A likelihood ratio test was conducted to choose between the two successful models. The null hypothesis is that the mean of the underlying distribution of |Delta~ is zero; the alternative is that the mean is non-zero. These are equivalent to a null that the half-normal specification for |Delta~ is correct versus the alternative that the truncated normal model is correct. The calculated likelihood ratio statistic is 1.024 which is less than 3.842, the chi-square critical value for one degree of freedom and an |Alpha~ = .05. Hence, the null cannot be rejected; the half-normal specification for |Delta~ is correct. Consequently, the results for the normal/half-normal model are presented in Tables I and II whereas the results for the normal/truncated normal model are contained in appendix Tables AI and AII.

The estimated model shown in Table I is similar to many earnings functions found in the literature. Earnings rise at a decreasing rate with both age and time with one's employer, and the rate of return to schooling is comparable to other studies. The earnings adjustment variables all play expected roles. For instance, living in an urban area, being married or a household head, being male, and possessing greater wealth each increase a worker's earnings. Furthermore, most occupations appear to earn more than the reference group of farming, forestry, and fishing workers and no industry pays less than the reference industry, agriculture. Living in the central city lowers a worker's earnings below the earnings of those living elsewhere.

The chi-square statistic shows that there is a relationship between the regressor set and the dependent variable. The pseudo-|R.sup.2~ is .32. This is on a par with most other cross-sectional wage studies. The reservation wage results found at the bottom of Table I indicate that the average worker earns about 25% per hour more than his/her reservation wage of $7.20. The bulk of reservation wages lie between $1.75 per hour and $12.65 per hour.

The main focus of this study involves the various hypotheses discussed earlier. Table II reiterates the hypotheses and shows the test results corresponding to each. Except for Hypothesis TABULAR DATA OMITTED 2, the results strongly confirm each hypothesis, suggesting that the proposed reservation wage measure is plausible. Each hypothesis test will be discussed separately.

Hypothesis 1 proposes an inverse relationship between rates of time preference and reservation wages. Thus married, prime-age males, assuming they have comparatively low rates of time preference, should set higher reservation wages than either females or young males. Table II reveals in both cases that prime-age male reservation wages are significantly higher.

Hypothesis 2 suggests that unemployment benefits, because they reduce marginal search costs, will be positively related to reservation wages. This hypothesis is not borne out by the test, however. The mean reservation wage of the second highest benefit category (40.2%-43.2%) is significantly greater than that of only two other categories: 33.2%-36% and 37.1%-40%.

Hypothesis 3 speculates that there is a direct relationship between education and reservation wages. We find that college-educated workers do indeed exhibit the highest reservation wages, whereas the least educated group sets the lowest reservation wages. Furthermore, reservation wages fall uniformly as education falls. These mean wages are significantly higher for those with college than for those who graduated from high school, attended high school, or never attended high school.

The test results for the remaining hypotheses are also convincing. Hypothesis 4 postulates that reservation wages will fall as search costs rise. Differential search costs are proxied by two indices of rural-urban living: SMSA vs. non-SMSA, and farm residence vs. nonfarm residence. In both cases urban dwellers set significantly higher reservation wages than rural workers.

Hypothesis 5 proposes that greater wealth lowers search costs, leading to higher reservation wages. Home ownership proxies wealth status, based upon the presumption that some wealth is necessary for home ownership whereas none is necessary for renting. Again, the results agree with the hypothesis. Greater wealth is associated with higher reservation wages.

VI. Conclusion

Using the stochastic frontier regression technique, this paper provides estimates of reservation wages of workers drawn from the Current Population Survey. Our results are novel in that we estimate reservation wages based on the observed wage structure, therefore avoiding possible sources of response bias arising in surveys that directly ask unemployed workers what their reservation wages are and in the sense that we have employed a data base that is more representative of the U.S. labor force than data bases used in previous studies.

We find the typical worker's wage is 25% higher than his reservation wage, which is consistent with the notion that most workers turn up job offers paying above the reservation wage. In addition, we test several hypotheses from search theory regarding reservation wages. Our results largely confirm these hypotheses. We find men have higher reservation wages than women and that reservation wages are directly related to age, education, labor market density, and wealth.

Finally, we find no systematic relationship between the generosity of a state's unemployment insurance system and reservation wages. This non-result may stem from the fact that workers in our sample by and large obtained work without taking the state UI replacement ratio into account in their job search. This would be true for most workers finding employment from new entrant, re-entrant, or quit status. Only workers permanently laid off or who chose to go to a new job from a temporary lay off would have reason to directly incorporate the level of the state UI replacement ratio into their search behavior. These workers likely make up a small fraction of the sample and therefore it's not surprising that their behavior fails to drive the results. Ideally, one would like to be able to distinguish among workers on the basis of the way in which they find employment in order to determine what effect if any the UI replacement ratio has on reservation wages. Unfortunately, this information is lacking in the CPS data base.

Data Appendix

All of the data used in the empirical analysis (with the exception of URATE83 and UI83) were drawn from the January 1983 Current Population Survey tape. The variables used in the analysis are defined as follows:

1. AGE is the worker's age.

2. AGESQ is the AGE squared.

3. EDATTEND is the highest grade attended by the worker.

4. EDAGE is education times age.

5. TEN is the number of years the worker has been employed by his or her current employer.

6. TENSQ is TEN squared.

7. SMSAYN equals 1 if the worker lives in an SMSA and is 0 otherwise.

8. LANDUSE equals 1 if the worker lives on a farm and is 0 otherwise.

9. MARSTAT equals 1 if the worker is married and is 0 otherwise.

10. RELHEAD equals 1 if the worker is the head of the household.

11. CENCITY equals 1 if the worker lives in a central city and is 0 otherwise.

12. URATE83 is the unemployment rate in the worker's state of residence. This variable was drawn from Employment and Earnings and matched to the worker.

13. SEX equals 1 if the worker is female and 0 if the worker is male.

14. OCCMP equals 1 if the worker is a manager or a professional and is 0 otherwise.

15. OCCSTCH equals 1 if the worker is a service, technical, sales, or administrative support worker and is 0 otherwise.

16. OCCCRFT equals 1 if the worker is a craft worker and is 0 otherwise.

17. OCCOPER equals 1 if the worker is an operative and is 0 otherwise.

18. OCCUNSK equals 1 if the worker is unskilled and is 0 otherwise.

19. INDMIN equals 1 if the worker is in mining and is 0 otherwise.

20. INDCON equals 1 if the worker is in construction and is 0 otherwise.

21. INDDUR equals 1 if the worker is in durable goods manufacturing and is 0 otherwise.

22. INDNON equals 1 if the worker is in non-durable goods manufacturing and is 0 otherwise.

23. INDTPU equals 1 if the worker is in transportation or public utilities and is 0 otherwise.

24. INDTRD equals 1 if the worker is in wholesale and retail trade and is 0 otherwise.

25. INDFIRE equals 1 if the worker is in finance, insurance, or real estate and is 0 otherwise.

26. INDSERV equals 1 if the worker is in the service industry and is 0 otherwise.

28. REGNE equals 1 if the worker lives in the Northeast census region and is 0 otherwise.

29. REGNC equals 1 if the worker lives in the Northcentral census region and is 0 otherwise.

30. REGS equals 1 if the worker lives in the South census region and is 0 otherwise.

31. NRKIDS is the number of children under 14 years of age in the worker's household.

32. FAMINC is the income of the worker's household net of the worker's own income.

33. OWN equals 1 if the worker is a homeowner and is 0 otherwise.

34. UI83 is the ratio of average weekly state unemployment insurance benefits to average weekly earnings in the worker's state of residence. This variable was drawn from the Social Security Bulletin and was matched to the worker.

TABULAR DATA OMITTED

TABULAR DATA OMITTED

1. Of course, because we focus strictly on workers who are employed, our sample may also be criticized as unrepresentative. In particular, because unemployed workers are omitted from the analysis, it may be the case that we are looking at workers with relatively lower reservation wages and ignoring workers with relatively higher reservation wages. If unemployed workers do indeed have higher reservation wages than employed workers possessing comparable observed characteristics, then our reservation wage estimates will be biased downward. Because the employed have been ignored in past studies of this issue and because the employed are more representative of the typical labor market participant, we would argue that this group is worthy of study. We issue the same caveat as Kiefer and Neumann, however.

2. See Schmidt |26~ and Bauer |2~ for comprehensive reviews of the technique's application to the problem of estimating production functions.

3. It seems clear that, due to the close connection between employer ignorance measures and reservation wage measures, this current work highlights the validity of the previous ignorance research.

4. See Mortensen |24~ for an excellent survey.

5. The normality assumption usually made with the use of t-tests does not hold in cases of one-sided error estimation. The t-test may nevertheless be applied for large samples since such samples permit considering the sampling distribution of the difference of each pair of means to be asymptotically normal |11~.

References

1. Aigner, Dennis, C. A. Knox Lovell, and Peter Schmidt, "Formulation and Estimation of Stochastic Frontier Production Function Models." Journal of Econometrics, July 1977, 21-38.

2. Bauer, Paul W., "Recent Developments in the Econometric Estimation of Frontiers." Journal of Econometrics, October/November 1990, 39-56.

3. Cowing, T., D. Reifschneider, and R. Stevenson. "A Comparison of Alternative Frontier Cost Function Specifications," in Developments in Econometric Analysis of Productivity: Measuring and Modelling Issues, edited by A. Dogramci. Boston: Kluwer-Nijhoff, 1983.

4. Crosslin, R. L. and D. W. Stevens, "The Asking Wage-Duration of Unemployment Relation Revisited." Southern Economic Journal, January 1977, 1298-302.

5. Feldstein, Martin and James Poterba, "Unemployment Insurance and Reservation Wages." Journal of Public Economics, February/March 1984, 141-67.

6. Fishe, Raymond P. H., "Unemployment Insurance and the Reservation Wages of the Unemployed." The Review of Economics and Statistics, February 1982, 12-17.

7. Heckman, James J., "Sample Selection Bias as a Specification Error." Econometrica, January 1979, 153-62.

8. Heywood, John S. and Sammis B. White, "Reservation Wages and Unemployment m Manufacturing: A Case Study." Applied Economics, March 1990, 403-14.

9. Hofler, Richard A. and Solomon W. Polachek. "Ignorance in the Labor Market: A New Approach for Measuring Information Content," in Proceedings of the 1982 Business and Economic Statistics Section, American Statistical Association, 1982, 422-25.

10. ----- and -----, "A New Approach for Measuring Wage Ignorance in the Labor Market." The Journal of Economics and Business, August 1985, 267-76.

11. Hogg, Robert V. and Allen T. Craig. Introduction to Mathematical Statistics, 4th ed. New York: Macmillan Publishing Co., Inc., 1978.

12. Holzer, Harry J., "Reservation Wages and their Labor Market Effects for Black and White Male Youth." The Journal of Human Resources, Spring 1986, 157-77.

13. Jondrow, James, C. A. Knox Lovell, Ivan Materov, and Peter Schmidt, "On the Estimation of Technical Inefficiency in the Stochastic Production Function Model." Journal of Econometrics, August 1982, 233-38.

14. Jones, Stephen R. G., "The Relationship Between Unemployment Spells and Reservation Wages as a Test of Search Theory." The Quarterly Journal of Economics, November 1988, 742-65.

15. -----, "Reservation Wages and the Cost of Unemployment." Economica, May 1989, 225-46.

16. Kiefer, Nicholas M. and George R. Neumann. "Estimation of Wage Offer Distributions and Reservation Wages," in Studies in the Economics of Search, edited by Steven A. Lippman and John J. McCall. New York: North-Holland Publishing Company, 1979, pp. 171-89.

17. ----- and -----, "An Empirical Job-Search Model with a Test of the Constant Reservation-Wage Hypothesis." Journal of Political Economy, February 1979, 89-107.

18. ----- and -----, "Individual Effects in a Nonlinear Model: Explicit Treatment of Heterogeneity in the Empirical Job-Search Model." Econometrica, July 1981, 965-79.

19. Lancaster, Tony, "Simultaneous Equations Models in Applied Search Theory." Journal of Econometrics, April 1985, 113-26.

20. ----- and Andrew Chesher, "An Econometric Analysis of Reservation Wages." Econometrica, November 1983, 1661-76.

21. Lippman, Steven A. and John J. McCall, "The Economics of Job Search: A Survey." Economic Inquiry, September 1976, 347-68.

22. McCall, John J., "Economics of Information and Job Search." Quarterly Journal of Economics, February 1970, 113-26.

23. Mortensen, Dale T., "Job Search, the Duration of Unemployment and the Phillips Curve." American Economic Review, December 1970, 847-62.

24. -----. "Job Search and Labor Market Analysis," in Handbook of Labor Economics, Vol. II, edited by Orley Ashenfelter and Richard Layard. Amsterdam: Elsevier Science Publishers, 1986, pp. 849-919.

25. Polachek, Solomon W. and Bong J. Yoon, "A Two-Tiered Earnings Frontier Estimation of Employer and Employee Information in the Labor Market." The Review of Economics and Statistics, May 1987, 296-302.

26. Schmidt, Peter, "Frontier Production Functions." Econometric Reviews, 4, No. 2, 1985-86, 289-328.