Does School Quality Matter? Evidence from the National Longitudinal Survey for Youth

I. Introduction

In business, the general complain is having to find high school graduate lacking basic skills which negatively affects the firm's competitiveness. In addition, policy analysts and educators have observed declining Scholastic Aptitude Test scores, with American pupils’ performance placed near or exactly at the bottom in international tests of academic achievement despite the increasing spending on their public schools over the last quarter century.

In previous studies such as Hanushek (1986, 1989), Coleman Report (Coleman et al., 1966), Card and Krueger (1992), eight studies measured the school quality at the state level, while seven measured the quality averaged by school district or county. Such approach is likely to create both aggregation bias and measurement error which are also particularly acute to state level. Measurement errors also arise from the assumption that a person is educated in the state of birth. Moreover, the use of state level data faces potential bias from unmeasured changes in other state characteristics. It is noticeable that among the published paper in this literature on school quality and earnings, none has measured the quality of the school at the level of the school actually attended. Furthermore, the measure of “quality” is in terms of spending.

The National Longitudinal Survey of Youth (NLSY) is used to test for a connection between the quality of the high school attended and the log weekly salary of white males. Usage of the data presents five distinct advantages.

1. There is a potential elimination of measurement error and aggregation bias which was observed in the previous studies.

2. Data set offers information on detailed school characteristics in contrast to overall expenditures.

3. By using micro data there is a control for individual characteristics.

4. Test whether the link between school quality and earnings reported for white males in Card and Krueger (1992a) study still persists today.

5. Using micro data along with a one-step estimator, school quality can be included as having a direct impact on each worker’s years of education.

The literature observes that though there is significant differences between the labor-market performance of students enrolled in different schools, these differences do not significantly relate to the standard measures of the quality of the school, which is in agreement with the literature on school quality and test scores as assessed by Hanushek.

II. Data

Source of the data is Geocode version of the National Longitudinal Survey of Youth (NLSY), 1979-1990. Data follows a set of employees aged 14 to 21 in 1979 and were 24 to 31 in 1989.

Sample includes observations on:

1. labor-market outcomes from 1979-1989 since the years of education which are important conditioning variables are not available on an annual basis in earlier years

2. only for workers aged 17 or higher in the given year

Sample excludes observations:

1. having missing data on earnings or collective weeks worked

2. having missing data on cumulative weeks worked

3. employees whose reported education decreased from one year to the next

4. employees in the military subsample – wherein no data on school quality were gathered

Thus the outstanding subsample comprises of 6,749 people with 68,182 observations. Also some younger pupils who did not go to high school included in the initial sample of schools, have lacking data which the loss of data may have been 4% of the sample.

The sample observations were further reduced to:

-  19,534: white males

- 18,395: attendees of public high school

- 17,706: living in the 50 states or the District of Columbia in the current year

- 16,417: having positive weekly earnings in the given year

- 16,406: with unbalanced panel of 1,812 individuals. This is because in regressions in which the teacher-student ratio was included as a regressor, the one person whose school reported a teacher-pupil of 6 (which on the contrary, no other school reported a ratio above 0.2) was excluded.

In the regression sample in May of the given year, the composition of enrollment status of the worker is as follows:

- 63.27% were high school graduates (or higher) not attending college

- 14.54% were high school dropouts

- 14.31% were attending college

- 4.86% were attending high school but graduated that spring

- 3% were attending high school and did not graduate that spring

The endogenous variable is the log of weekly earnings derived by dividing annual earnings by weeks worked. However, regression using hourly wages and yearly earnings as the dependent variable yield very similar results.

Three variables were used as the principal indicators of school quality:

- ratio of full-time equivalent teachers to students

- salary of starting certified teachers with a B.A. degree and lastly,

- percentage of teachers with a Master’s degree or higher

The beginning salary is converted to a relative wage by dividing it by mean earning per capita in the state, variable acquired from the Regional Economic Information System of the Bureau of Economic Analysis. Approximately 550 schools are represented in the regression.

The available information on the percentage of teachers with Master’s degree, data class size and teachers’ wages were 78.5%, 75.4% and 77.4% of the sample of white males, respectively.

The general sample size for a regression given the three measures of school inputs is 11,314.

III. Results

A. Do Schools Differ?

With everything constant, it would be too early to look for school attributes which influence earnings without first determining whether earnings vary between employees attended different high schools.

The null hypothesis is that schools have no effect on earnings. The output from the models in the study suggests that high school attended does significantly influence the earnings. Furthermore, even if personal and family attributes were included in the regressions used in the paper the null hypothesis is still rejected. Hence it shows that schools do vary considerably in quality.

B. The impact of School Quality on Earnings

The measures of school quality used:

- the ratio of teachers to pupils,

- the wage of beginning teachers in relation to earnings per capita in the state

- fraction of teachers with Master’s degree or higher

However, none of the three school quality variables is considerably related to salary. Moreover, the elasticities at the means of weekly earnings regarding school quality are very little.

One potential explanation on the insignificance of the three variables may be due to their collinearity with each other. To verify this possibility the model is repeated using one quality variable at a time – however the result is still consistent: the relation between school quality and log weekly earnings of the employees is statistically insignificant. 

C. Diminishing Returns to School Quality

The three quality variables are insignificant probably because many of the salary observations include temporary jobs and summer jobs taken by college students. These jobs may have little relation to the “career” job for which the person is preparing for.

Geographic mobility may also expound on why employee’s earnings have no relation to the examined quality of his high school. However the data of the student’s length of stay in that school is unavailable.  The idea is approximately tested by doing a basic regression on the subsample of employees who had transferred to the city where the school is located – the result shows insignificance of the school quality.

Furthermore, the measurement error might have diminished the estimated effect of attending a good school on earnings given the likely flaw of the data set which is from the fact that the information on the schools was gathered in 1979, long after some of the employees in the NLSY had exited secondary school. Considering this idea, the school variables remain not significant.

In Farber and Gibbons (1991) argument, for the new members of the labor force, there is found upon hiring to be little but unknown correlation between earnings and any measurement of employee efficiency. Nevertheless as the firm learns more about its worker it would soon adjust its salary to match the employee’s marginal efficiency which in turns increases the correlation between earning and unobserved measures of productivity – thus further explaining why the results based on employees aged 31 to 60 (at the time of data collection) differ from the present results based on employees aged 17 to 32 (during the course of the survey).

Finally, with the exclusion of those having less than 12 years of education (on the basis that those dropouts may not respond to school quality at the same degree as high school graduates, because of unobserved heterogeneity) the relative teacher wage variables are significant  in the regression. However it is implied that higher teacher wages are linked with lower earnings for those student having less than 15.4 years of schooling.

In conclusion, regardless of repeated regression under different specifications, and on different subsamples in the goal of eliminating potential data problems same output is still derived: the relation of the three normally used measure of school quality has no positive significance to the later earnings of pupils.

D. The Possibility of Endogenous Regressors

1. Does Attendance at Better High School Allow Students to Acquire More Education and Work Experience?

So far education and experience have been treated as exogenous regressors. However if first-rate schools include earnings exclusively by making their pupils acquire more education and work experience (as compared to low-rate schools), hence the coefficients on the school quality variables would be biased by the addition of education and experience in the earnings equation.

In summary of the regressions done, the results do not provide much support to the belief that better high schools contribute to earnings by inducing employees to acquire more educations and/or training than what they would otherwise get.

2. Endogeneity of the School Quality Variables:

The quality of the high school attended by a pupil may likely be an endogenous function of his and his family’s attributes, such as children are more inclined to go to a good school given parents of high socioeconomic status. This possibility does not seem to apply in the present data set given that such endogeneity would potentially force an upward bias on the coefficient on school inputs, presuming that the unmeasured socioeconomic status of parents is positively associated with earnings.

Examining this idea further, the correlation coefficients between the school quality variables and five measures of family background (father’s and mother’s education, family member had a library card at age 14, student in 1979 still in city of birth, and imputed wage of father) were derived. The highest correlation was observed to be 0.255 between the percentage of teachers with master’s degree or higher and imputed wage of father.

Next, Hausmen test for the endogeneigty of the three school quality variables and three interactions among the variables and years of education. The result shows that the endogeneity of school attributes does not present a significant problem. The lack of a significant contingence of school quality on family background may not be the expected outcome to some, however recall that the study is exclusive to employees who attended public schools and it might be the case that those parents who become concerned in selecting schools do so by enrolling their children in private schools.

E. Other School Characteristics

Since the paper found no significant relations between the three school quality variables and the later earnings of workers, and the three measures also do not predict student’s earnings, the study is concerned if there are alternative school attributes.

Out of the seven additional school characteristics chosen four were found insignificant in any of the regression used:

1. library books per student enrolled

2. a dummy variable set to 1 if any of the 7 vocational curricula were accessible at the school

3. the percentage of black in the student body

4. the percentage of teacher who had exited the school in the preceding year (excluding reasons of death and retirement)

The three additional school attributes found to be significant in relation to the later earnings of pupils are:

1. number of enrollees at the school (positively significant)

2. Schools having a higher proportion of disadvantaged students are more likely to have alumnae with lower earnings, at least for those having attained less than 12 years of education. Consequently, the proportion of grade 10 students who drop out without finishing grade 12 has a negative and considerable impact on earnings, at least for those employees who have less than 13.4 years of education.

3. school size as measured by enrollment – which is also found to be the only measure of school “quality” that is observed to be significantly related to pupil’s subsequent earnings

IV. Potential Explanations for Why School Quality Does not Affect Wages

A. The Impact of State Averages of School Quality on Earnings

For this case the earnings regression is repeated using state-level measures of school quality rather than data on individual schools from the NLSY’s 1979 school survey. A total of 12 regressions are made with the first six having included only the state-level measure of quality and the last six include the true measures of quality for the actual high school attended. The t-statistics of the first six are generally higher than that of the latter.

One explanation of the result is that the state-level measures of school quality are seizing some feature of the state other than the quality of the actual high school attended. It may be likely that in contrast of high school quality, school quality in the lower grades has no impact on pupils’ earnings – an effect seized by the state averages. Regrettably, it is impossible to test the idea on the NLSY data.

B. Has School Quality Converged?

In previous analyses such as that of Card and Krueger (1992a) it was stated that over the last few decades, school quality has come together both within and between states, eliminating the major wage discrepancies emerging from the differences in school quality. Hence the data used by Card and Krueger might capture true school quality effects presented on workers who attended schools some decades ago, whereas the NLSY shows little (if any) impact, given that by 1979 common measures of school quality had converged substantially.

One result implies that the convergence was specifically strong for teachers’ mean salaries. However, there is still considerable disparity in measured quality across the individual schools in 1979. There is also an observed small dispersion of teachers’ relative starting wages which may justify why the variable can explain so little of the dispersion in student’s later earnings.

The teacher-pupil ratio seems to have substantially more dispersion than the relative salary variable, whereas at the other side is the proportion of teachers with Master’s degrees or higher which is highly spread.

The most conservative conclusion that can be derived from the analysis is that there is a very small deviation in teachers’ relative salaries, and such deviation is insignificantly related to the variation in the pupils’ subsequent earnings, and that the likelihood remains that having sharper salary differential between schools – statistically significant effects on student outcomes would arise.

C. Diminishing Returns to School Quality?

From the previous analysis, it was observed that the mean school quality at the state level increased sharply between 1939 and 1979, thus the marginal benefit is expected to be larger from a rise in the quality of the school in the early part of the century as compared to the present.

Taking this idea, regressions are done to test for whether there are still considerable returns to school quality once non-linearity such as those due to diminishing returns are considered. However none of the school quality variables was found to be significant.

D. Has There Been a Structural Shift in the Link between Schooling Inputs and Schooling Outputs?

It may be that the contrasting results of Card and Krueger’s study with those of this paper represent the true education production function in the first and latter halves of this century, respectively. In order to resolve the two sets of outcomes, the type of structural shift in American public education must be considered.

Two hypotheses for the cause of such a shift:

1. The increase in bureaucratization of public school may have undermined the connection between standard controls of school quality and educational outcomes. In other words, the increase diminishes the public schools effectiveness.

2. Large number of court decisions over the last three decades in pursuit of removing gaps between school districts is the structural shift that might have weakened the connection between the standard measures of school quality and student performance over time.

V. Conclusions

The study has observed that income of white male employees relies extensively on which high school they have attended. On the other hand, standards of school quality explain very little of the differentiations between schools. Such finding is found to be highly robust to alterations in specifications in the sample of the study. However, tests for endogenous regressors showed no evidence of such problem.

Moreover, no oddity was found in the NLSY data set. Previous study using the same data, (in comparison with this current study) also found various measures of school quality had either insignificant impacts or effects varying in sign between the data sets used.

In addition, the explanations for the low correlation between standard measures of high school quality and student subsequent income are as follows:

1. Lack of variation in the data may justify to some degree the results for teacher’s wages, however it cannot explain other findings

2. Possibility of diminishing returns was rejected

3. Structural shift has altered the link between  school inputs and earnings over the last several decades

As a result, though high spending might be important to enhance student performance it may not in itself be enough. One likely reason for the observed inefficiencies is the flaws of the incentive structure of the administrators, teachers and students.

It seems that there is a need for more subtle measures of school and teacher aspects, and probably of incentives so as to determine policies that may improve the later earnings of high school alumnae.

Source:

Julian R. Betts, “Does School Quality Matter? Evidence from the National Longitudinal Survey for Youth”, Review of Economics and Statistics. Vol. 77, No. 2, (May 1995), pp. 231-250.