PRIVATE EDUCATION AND THE SIGNALING EFFECT

PRIVATE EDUCATION AND

THE SIGNALING EFFECT

M

T

IE&B

Faculty of Economics & Business University of Groningen

2

ABSTRACT

In this paper, I explore the relationship between ability signaling and the private or public nature of one’s education choice. Following a methodology developed by Altonji and Pierret (2001), I find evidence that the central mechanisms of education within the labor market function differently now than in the past. Furthermore, I do not find evidence that either public or private education function as a signal for innate graduate ability. Finally, I propose a reassessment of the central contributions of education within the labor market, in relation to the modern economy.

3

INTRODUCTION

Ever since Schultz’s 1961 article on human capital and productivity, a lot has been written and researched on the topic of education. The contributions of educational attainment are of great importance, both for the economy as a whole and for individual persons. The idea that education adds to the productivity of workers is not a new one; training and learning teaches employees skills, which in turn allow those workers to contribute more to the economy due to their greater efficiency in performing their tasks. Economy-wide productivity is determined not only by the combined productivity of all workers, but also by the efficiency of allocation of labor. Not only should workers be allocated jobs which suit their skills and traits, the most productive workers should be allocated to position where they add the most value to the global economy. In a perfect economy, this would correspond to high wages returns to the most skilled, motivated, healthy employees, among other traits. Note, before I continue, that although I refer to both universities and colleges in this paper, these are interchangeable for the sake of this article. Within the scope of this paper, I consider different types of tertiary education similar enough to group them in the same category of post-highschool education.

In 1973, a new aspect was added to this theory of educational returns. Michael Spence (1973) introduced his model of job-market signaling, where he argued that workers are willing to make investments (i.e. in their formal education) to signal their skills to employers, if this results in a higher expected wage. The employer, in turn, pays higher wages to credible signals of capability such as an educational background, as the conveyors of these signals are generally more productive than uneducated employees. A particular example of such a signal is the choice in educational institution by prospective students. Prestigious universities tend to charge significantly higher tuition fees than common ones, but the flipside is that prospective students might believe education at those expensive institutions brings rewards in the long run, be it from higher expected future wages or from prestige and family tradition. Though there are certainly prestigious public universities, it is commonly understood that the most esteemed universities around the world are private universities in the American sense: educational institutions that rely on student and alumni funding rather than government grants to operate. These universities, such as Harvard, Oxford, Cambridge and many others around the world, are generally much more selective in acceptation applications and charge tuition fees approximately 10 times that of the average public university. The resulting exclusivity and investment by the students creates an opportunity for a strong signaling effect, which is the theoretical basis behind this paper.

4 and have decent communication skills, among other signals. Later frameworks of ability signaling were built on the theory that employers gain information on the ability of their employees over time, so that over time they rely increasingly less on education as a signal. However, Arcidiacono et al. (2010) found that though this holds true for high school graduates, college degrees remain a good indicator of ability over time regardless of experience. Fang (2006) tried to measure these separate effects of education, and found that only two-thirds of the college wage premium could be attributed to the enhanced productivity, meaning that signaling amounts to a third of the wage gain from education. Research in this area is of increasing interest to the field of international economics, as insights in the workings of the labor markets and efficiency of (educational) institutions are key to both policy makers and employers. Policy makers will want to have accurate information on causes and effects in the economic system, such as the determinants of labor productivity. Whether the goal is greater individual productivity or a more efficient allocation of labor, research in this area leads us to develop better economic policy. In particular, many countries have a dual educational governance system with both public and private universities, and there is an important role for the government to balance both legislation and funding for both types, to produce the best outcome for the economy as a whole. The international importance is fueled by for example the integration of the European Union, of which the open labor market is a key feature. Increasing and diversifying international student flows (Bhandari & Blumenthal, 2011) are another indicator that the educational environment is a global one and as such deserves a place in the analysis of international economics. International employers will be interested as well in the functioning of different educational structures. Generally a human resources department might be interested in the different contributions a private university makes, for example, over a public university, and how this plays a role in the employee’s productivity.

CONTRIBUTION OF THIS PAPER

5 First, University graduates are more productive than high school graduates due to actual skills learned in the degree program. All things equal, those who have attended university will at least have learned some skills, either active or passive, that are usable in their prospective careers. Even those who eventually end up in careers that do not match their field of study will often end up using techniques or thought processes in their working life. However, there might be a difference between private and public university degrees, though there is no conclusive evidence on the educational quality difference. Seneca and Taussig (1987) for example find that there is no direct link between tuition and quality of education. Rather, public universities suffer from state requirements to be accessible to a wider range of students, which leads to a quality disadvantage for most public universities. As such, public universities perform worse on average than private universities, but there are definitely exceptions to this rule. They argue that the quality difference between the few to public universities and most private universities can be instead attributed to size differences: smaller public universities have less necessity to cater to masses and are very able to compete with private universities on the quality rankings.

Second, there are the signals that come with university education in general. It is not unthinkable that the people who choose to acquire a university degree are not a random sample of the population, but are in some form a ‘crème of the crop.’ These might include better communication skills, sufficient motivation to complete a multiple-year study program and other characteristics. Note that this does not mean all university graduates have these characteristics, but rather that on average they are expected by employers to have so. This is the traditional signaling effect. With respect to the investment mentioned in the signaling theory, takes into account both time and money invested in any university degree, as well as opportunity costs from not working a job instead of studying.

Finally, there is the signal that is specific to graduates from private universities. By committing themselves to large investments (average tuition for private colleges was approximately $29.000 yearly in 2013, versus around $8.600 for public four-year colleges1_),

employees can signal that they believe they possess qualities such that they will later repay this with a better job. Following signaling theory, they might signal their ambition combined with their suitability for the highest skilled jobs. If employers find this suitability signal credible, then we might indeed find that there exists a wage premium on private, high tuition education that is not due to college-taught skills or general university student abilities. The goal of my thesis is then to examine whether there is indeed a difference in the wage premium build-up between public and private university graduates. This wage premium consists on the one hand of the actual skills learned during one’s educational history, and on the other hand of an expected set of abilities that is based on indirect sources such as a

6 person’s choice of university, which is not for certain related to those abilities. The central research question is as follows:

 Is there an additional signaling effect in the wage build-up of private university-schooled workers?

The existence of a tuition-based signaling effect is not only relevant for employees and employers, but also for policy makers for the whole economy. This tuition signaling can provide another sorting mechanism to improve the human capital allocation within the economy. This is especially true now that the main source of economic growth shifts from production to services (van Ark, O'Mahony, & Timmer, 2008), leading to an increased role for reallocation of labor in economic growth. Additionally, the theory of ability signaling is important in an international context, for example due to the widespread student exchange programs. More knowledge on the mechanics of educational choices will in turn provide us with knowledge of international labor markets and the flow of human capital across borders.

LITERATURE REVIEW

The modern economic literature that tries to prove and measure the signaling effect of college education in general starts with Farber and Gibbons’ (1996) model of ability learning. They develop a dynamic model of employer learning (or public learning) to study the wage effects over time of both innate ability and education. Their most important finding in relation to the research question of this paper is that the correlation between wages and unobserved ability (to the employer) increases with experience. The implication is thus that there exists a learning effect in which employees reveal their true abilities to the market over time. Another finding by Farber and Gibbons is that the relationship between education and wage remains positive as experience grows. Finally, they find mixed evidence that the wage residuals of their model are a martingale process.

7 unobserved by the researcher but not by the employer, which could have an accidental correlation with the observed ability (education).

Related literature also includes research on college graduate income development, such as by Thomas (2000), who finds that income growth patterns in the years after graduation are affected by the type of institution, the graduate’s major and the graduate’s performance. Later literature investigates other effects of degree characteristics, such as income growth patterns (Thomas & Zhang, 2005) and student access and choices (Clarke, 2007). Although these articles do not consider the background of income changes in terms of ability origins and signals, they do highlight the importance of characteristics such as institutional type and quality. However, there is little empirical testing in this field aside from the work of AP. Aside from the employer learning strand of empirical research, others have taken approaches of group discrimination (Moro & Norman, 2004) and endogenous educational choices (Fang, 2006), both of which focus on task specialization of groups.

Very little research has however been done on the relationship between signaling and private education, even though the latter has long been an important focus in the literature surrounding education in general. For example, Vandenberghe and Robin (2004) use several methods to measure the returns to education across countries and find mixed results. In some countries, private education seems to perform significantly better than public education, whereas in other countries (specifically where private schools are mostly religion-governed) public education is consistently better at educating students. The global applicability of research on private education is emphasized by research such as by Heyneman and Stern (2014), who are among a growing group of researchers that emphasize the importance of private education to not only serve the rich, but also provide quality education to the poor in low-income countries. On a more macro-economic scale, Bräuninger and Vidal (2000) explore the growth effects of different government policies related to public and private educational institutions. They find that even though government funding for education increases the number of skilled employees and economic growth generally, there are scenarios where a partially public educational sector maximizes growth.

This literature combined raises the question of whether there is a place for private education in the literature on employer learning. If the literature shows that in some areas, private education is better than private education, then it is not unthinkable that this sentiment is also felt by employers. Then if employers translate their knowledge of the advantages and disadvantages of private education into favorable hiring practices towards private university graduates, an opening is created for ability signaling through this private education, if it is indeed believed that graduates from private universities are biased towards being more skilled than public university graduates.

HYPOTHESES

8 H1. There exists a positive additional signaling effect on wages that is unique to the

private nature of the attended university.

This hypothesis follows from the original signaling effect as identified by Spence (1973), which stems from the investment those employees make, and the fact that employers find this signal to be credible. Although there are exceptions to this rule, private educational institutions are in general more expensive than public institutions, at least in the United States, because government-funded schools are required to be accessible to poor students. Though there are a number of public universities with higher tuition and rankings similar to top-tier private universities, I do not believe this would influence the credibility and value of the ability signal in general. Additionally, I formulate two additional hypotheses:

H2. There exists a positive general signaling effect on wages for all university graduates. H3. There exists an overall positive effect of education on wages, consistent with

productivity gains.

For hypothesis 2, I expect to also find a general positive signaling effect for all university graduates that is separate from the productivity gain from the actual skills learned in the educational process. This is in line with the findings from AP and other economists such as Fang (2006). University graduates have made investments in their educational development, both in the form of tuition and in the form of opportunity costs, and I hypothesize that university graduates are a biased sample with regards to skill, in that university graduates are generally a more skilled sample than the general population.

Finally, I expect to find a positive effect of education on wages, consistent with traditional wage theory that increased productivity is rewarded with a higher income reward. Again this is consistent with previous findings from empirical research on breaking down wage gains. Most jobs make at least minimal use of skills learned in the prerequisite education, but in general higher paid jobs require specialized skills that can only be learned from specific education and cannot efficiently be learned on the job.

THEORETICAL MODEL

9 these are revealed to the employer through evaluation of the employee’s performance history during employment. Newly graduated job applicants generally have a hard time communicating these skills to their prospective employers, as there are hardly any occupational sectors that have a standardized test. A notable exception is the military, which measures applicant suitability for military positions by their Armed Forces Qualification Test (AFQT) scores. Other commonly reported tests by applicants such as ACT, SAT and LSAT scores are college application tests rather than job skill tests, and as such are hardly valuable to employers as a source of information of an individual employee’s hidden abilities. The third variable in this model is experience, 𝑥, which reflects the exposure the employer has had to the innate skillset of the employee. Through interactions between the employees and their supervisors and through evaluation of employees’ job performance, employers gain information over time on the functional skills of employees.

The existence of a signaling effect of education is then observed if s and z are correlated, so that education is informative of the innate ability of the employee, and the coefficient of z will rise with experience and the coefficient of s will fall with experience. If that is the case, then as the employer is exposed to the employee’s job performance, he bases the wage choice to a larger extent on observed ability and to a lower extent on expected ability, inferred from schooling history. Additionally, if the coefficient of s remains positive for high values of 𝑥 , this indicates a productivity effect of education separate from the signaling effect, assuming a relatively competitive labor market where changes in productivity cause changes in wage levels.

Extending this model, a separate signaling effect of private education can be distinguished by introducing a variable p which indicates the individual’s history in private education. Theoretically this p is then a special form of s, in that it is a possible manifestation of innate ability that is observable to the employer. Then if the wage 𝑤 at 𝑥 = 0 is higher for private than for public institution graduates, and the coefficient of p declines faster with experience, we can deduct that employers consider private education a stronger signal of innate ability and discriminate as such when hiring employees, similar to the wage discrimination on years of schooling in AP’s model.

DATA AND METHODS

10 Wages are measured as the log of the hourly wage of the job with the most weeks worked each year, inflated for changes in price indices over the years. Hourly wages below $5 are dropped from the dataset, as they are below the legal minimum wage and thus might provide inaccurate data. For example, tipped jobs such as restaurant waiting staff will have a lower hourly rate than their actual hourly income. Furthermore, wages above $100 are set as $100, due to a few extreme outliers. Out of the 8687 subjects in the dataset, only 224 feature one or more years with an hourly wage above this threshold and for most subjects this is only one year. I start measuring income in the first year that the subject is not enrolled in any education, so as to filter out part-time student jobs, for example.

Military jobs are filtered out, and the subject is marked as unemployed for the duration of the military enrolment, due to the unique nature of military employment. As mentioned earlier, prospective soldiers are tested for ability much more thoroughly than common job applicants. Furthermore, military promotions have different determinants than normal job promotions, such as combat proficiency, to name one. As such, wage determination does not function necessarily as I assume in my model.

Additionally, I filter out self-employed jobs, because the concept of hidden ability in my context is not very relevant in that case. Self-employed individuals should be aware of their qualities to the point that if they are not, they will generally not gain this information by observing themselves work. Self-employed workers will also often set their wage as a portion of (expected) profits or even as the total residual income after paying for all company costs. However, I still consider their time worked as self-employed as job experience for future employers, as they can infer hidden employee skills from the performance of the self-employed’s company, for example.

I measure s as the highest educational grade completed by the subject at the current interview round. Subjects can re-enroll in education to increase their higher education, but I assume that someone’s educational attainment does not decrease. An example of this is that a person who completes two years of a four-year program and then starts a new program of the same level (i.e. both Bachelor programs) but at the first year again, this person is still counted as having two years of higher educational attainment. Values of s up until 12 correspond to the 12 elementary school and high school grades, values over 12 correspond to grades completed in university, including graduate programs.

AFQT scores, which I mentioned earlier, are used as a measure of z. Contrary to the NLSY79 data used by AP, the NLSY97 pre-standardizes the ASVAB test results to reflect the AFQT score percentile for the subjects’ age group.

11 interaction between employers and employees. There is a natural difference in work week length between occupations and it can be expected that employers adapt their review mechanisms to this. Weeks worked before graduation are also taken into account. Even though this might overestimate the experience of employees who had a lot of part-time jobs in their youth, information learned by employers might very well be carrier over between pre-graduation and post-pre-graduation employers. This holds especially true if employees stay with the same firm when they graduate.

I measure p, private educational attainment, as the number of grades someone has completed in private college education. This retains the same value if someone completes a higher public degree. Note that both for s and p, grades completed are taken into account even if the subject does not complete his degree. This is because in terms of signaling, partially completed degrees are still communicable to prospective employers through resumes for example, and skills learned during the first grades are also retained. Data for college type comes through the NLSY database from the IPEDS database and distinguishes between public, private for-profit and private non-profit institutions, where I merge the last two categories into one private college group. I only select male subjects of black and non-black/non-white ethnicity.

DATA DESCRIPTION

The main dataset consists of a panel of 8984 unique individuals, whom together group 134760 observations. Of these, the final regressions use panel data of 1427 individuals and 7174 yearly observations. A table of the main summary statistics can be found in appendix 1.

Hourly wages are heavily skewed toward the lower end, with a mean wage of $10.49 and a standard deviation of $8.02. Within a panel setting however, the standard deviation is much smaller: $5.76 between groups and $6.14 within groups. This, along with the existence of many outliers towards the higher end, emphasizes the need to convert the wage variable to a logarithmic scale, even with a wage cap of $100. Schooling levels after graduation are somewhat normally distributed, with spikes at 12th _{grade (end of high school, 53% of subjects)}

and 4th _{year of college (standard length of a four-year college degree, 23% of subjects).}

12 results in 1883 individuals reporting having completed one or more years of private university. Of these, 44% complete at least a 4-year degree.

MODEL DESCRIPTION AND ESTIMATION

In order to test my model, I run four random effects linear regressions using the data above, with standardized variables. One reason I do not use a fixed effects model as the Hausman test suggests for my data is that I have time-invariant data that has multicollinearity with the fixed effects estimator. Another reason is that I feel the data is a sufficiently random sample of the population to justify random effects. Regressions 1 and 2 are similar to those run by AP, and serve to test the validity of their model with the current dataset. The third and fourth regression are used to identify the private education-specific signaling effect. In these equations, 𝑤𝑖 is person 𝑖’s log real wage, 𝑠𝑖 is school grades completed, 𝑧𝑖 the standardized

AFQT test score, 𝑥𝑖 is weeks of working experience, and 𝑝𝑖 is the number of private university

grades completed.

(1)𝑤𝑖 = 𝛽0+ 𝛽1𝑠𝑖+ 𝛽2𝑧𝑖 + 𝛽3𝑠𝑖∗ 𝑥𝑖+ 𝛽4𝑥𝑖

(2) 𝑤_𝑖 = 𝛽₀+ 𝛽₁𝑠_𝑖+ 𝛽₂𝑧_𝑖 + 𝛽₃𝑠_𝑖∗ 𝑥_𝑖 + 𝛽₄𝑥_𝑖 + 𝛽₅𝑧_𝑖∗ 𝑥_𝑖

(3) 𝑤_𝑖 = 𝛽₀+ 𝛽₁𝑠_𝑖+ 𝛽₂𝑧_𝑖 + 𝛽₃𝑝_𝑖 + 𝛽₄𝑠_𝑖∗ 𝑥_𝑖 + 𝛽₅𝑝_𝑖∗ 𝑥_𝑖+ 𝛽₆𝑥_𝑖

(4) 𝑤𝑖 = 𝛽0+ 𝛽1𝑠𝑖+ 𝛽2𝑧𝑖 + 𝛽3𝑝𝑖 + 𝛽4𝑠𝑖∗ 𝑥𝑖 + 𝛽5𝑝𝑖∗ 𝑥𝑖+ 𝛽6𝑥𝑖 + 𝛽7𝑧𝑖 ∗ 𝑥𝑖

Contrary to AP’s regression models, experience is modeled linearly. I tested for different fits of time and experience with income, but quadratic and cubic polynomials did not provide a better fit than a simple linear function. All equations control for time trends, 𝑠𝑖 and 𝑧𝑖

interacted with a time trend, two-digit occupational code, and urban or rural residence. Equations 3 and 4 also control for 𝑝𝑖 interacted with a time trend. I considered using year

dummies instead of a time variable, but those year dummies proved statistically insignificant, and their inclusion did not improve the fit of other variables.

As a note, I do not consider endogeneity a particular threat within this model. Due to the timeliness of the variables in the model (individuals first obtain an education and then find a job that earns them a wage), there is little room for causality to run the wrong way, also for the other variables. High income from wages could fund a re-enrollment in (private) education, but the results which I will discuss in the next chapter does not seem to support such a causality.

Following the theoretical model of this article, the main prerequisite to identify a signaling effect is for 𝛽5 in (2) and 𝛽7 in (4) to be positive and significant, indicating that employers

learn about a worker’s skill over time. Then if 𝛽3 decreases between (1) and (2) and 𝛽1 is

13 (4) and 𝛽3 is positive, this would reinforce the main hypothesis that there is a positive

signaling effect of private education on wages. In that scenario, an employer bases his productivity estimate of the employee at 𝑡 = 0 on the signal from 𝑝𝑖, but as 𝑥𝑖 increases, the

employer learns the true skillset of the specific employee and alters his productivity estimate, and thus his wage offer. One important note is that the coefficients for 𝑠𝑖∗ 𝑥𝑖 and 𝑝𝑖∗ 𝑥𝑖 do

not have to be negative for this. There might be an inherent positive wage growth path for higher- and private educated employees. Also note that the control variables include a time trend interaction with 𝑧𝑖 in all four regressions, intended to control for these growth path

differences conditional on their inherent skills and abilities.

EMPIRICAL RESULTS

The estimated main coefficients from the four regressions are shown in table 1 below. More complete regression results, including control variables, are found in Appendix 2.

Table 1 (1) (2) (3) (4) 𝒔_𝒊 -0.0437** -0.0405** 0.0891 0.0882 (-2.86) (-2.64) (0.92) (0.91) 𝒛_𝒊 -0.0203 -0.0313* -0.123* -0.115* (-1.48) (-2.16) (-2.53) (-2.24) 𝒑_𝒊 -0.307*** -0.308*** (-3.77) (-3.78) 𝒙_𝒊 0.359*** 0.350*** 0.514*** 0.500*** (9.70) (9.43) (4.18) (3.97) 𝒔_𝒊∗ 𝒙_𝒊 -0.106** -0.0610 -0.278* -0.284* (-2.68) (-1.39) (-2.02) (-2.06) 𝒛_𝒊∗ 𝒙_𝒊 -0.0511* 0.0290 (-2.32) (0.52) 𝒑_𝒊∗ 𝒙_𝒊 0.0505 0.0439 (0.73) (0.62)

T statistics in parentheses. * p<0.05, ** p<0.01, *** p<0.001. [] ∗ 𝒙𝒊 indicate

interaction terms for experience. All coefficients are standardized betas.

The most important fundamental result for the evaluation of my hypothesis is lack of a positive coefficient of 𝑧𝑖 ∗ 𝑥𝑖 in regressions (2) and (4), signifying a lack of employer learning

of hidden abilities. The interaction of 𝑧𝑖 with time however (see appendix 2) is significantly

positive in all regressions, dispelling the possibility that inherent wage growth path differences for skill groups might influence these results. Innate ability seems to put employees on a faster wage growth path over time, but this seems to be unrelated to the signaling effect. Rather, skilled people might simply learn on the job fast over time than unskilled people.

I fail to find confirmation for the employer discrimination phenomenon that AP find in their research, where employers reward educated employees with higher wages and/or better jobs. Aside from the lack of skill learning, the coefficient for 𝑠𝑖 ∗ 𝑥𝑖 remains increases

14 negative coefficient that AP find, the conclusion is that education does not degrade with experience as an estimator of underlying skills. However, similar to the coefficient for ability, the interaction between time and education does increase over time. Combined with the negative coefficient for 𝑠𝑖, the conclusion is that higher educated graduates earn less after

graduation than lower educated individuals, but their wages increase faster with time, not with experience. This might indicate that productivity gains from education only reveal themselves over time. Furthermore, the individual coefficient for experience is positive in (1) and (2), whereas the coefficient for time is negative. Though the latter seems counterintuitive, it makes sense as we control for someone’s level of experience. It conforms to the idea that if you are unemployed for a longer period of time, you become less desirable in the labor market. The time coefficient then mostly corresponds to time passing while working below average weeks each year.

The lack of confirmation of the hypotheses continues in the third and four regressions, which test the hypothesis that college type serves as a signal for unobserved abilities. 𝑧𝑖 ∗ 𝑥𝑖 is

insignificantly different from zero, which indicates again that employers also do not use the private or public nature of the educational background as a signal for that hidden 𝑧𝑖.

Furthermore, there seems to be a wage benefit from private education within this dataset over time, but not immediately and not with experience. The coefficient for 𝑝𝑖 ∗ 𝑥𝑖 is not

significantly different from zero, whereas the coefficient for 𝑝𝑖 is negative and the coefficient

for the interaction with time is positive. It seems that the different effects of schooling in the first two regressions are actually caused by effects of private education alone, as the inclusion of the private education variable and interactions render the coefficients for schooling insignificant (except for 𝑠𝑖∗ 𝑥𝑖) and the coefficients for private education take on the signs

and significance of schooling in regressions (1) and (2).

Two findings that are consistent between the regressions are the coefficients for experience and the coefficient for ability. Contrary to one of my assumptions, there seems to be a strong positive relationship between work experience and wages. In the first regressions this is mitigated by a negative relationship between time and wages, but the second pair of regressions, which include private education, experience has a purely positive effect on the hourly wage. In fact, in regression (4), it is the only remaining determining factor besides two private education effects and the urban/rural control variable. This suggests strongly that learning on the job is the main determining factor in improving someone’s productivity. The second consistent finding is that the coefficient for ability ( 𝑧𝑖) is consistently negative

15

DISCUSSION

The main difference between AP’s findings and the results of this paper is that lack of employer learning found by me. Despite using a similar methodology with similar data, I cannot replicate the earlier findings and I cannot find confirmation for my own hypotheses. It seems as though there has been a structural change in the determinants of wages and the effects of education and experience in the past thirty years.

In finding an explanation for these observations, we can look both at the employee and the employer. Starting at the latter, one possibility is that the economic environment has changed between the timeframes of both articles. Reduced communication costs and increased availability of information has reduced the necessity for employers to rely on indirect information such as years of schooling or choice of university to estimate a person’s unknown skills. This has made it much easier for employers to verify an employee’s history, both work- and non-work-related. The rise of the internet in the past twenty years and in particular the rise of social media has made background checks much easier for even the smallest employers, who don’t have the time and money for conventional, professional background research. Additionally, cheaper communication and better communication methods has made training employees much more efficient and widespread, with the introduction of all kinds of electronic learning programs. This might mean that employers have become much less dependent on the knowledge and experience of employees beforehand, and subsequently cause the wage premium for knowledge and experience (or the interactions which we examined in the data analysis) to become less significant.

On the employee side of job finding, greater ease and lower cost of communication have also contributed to better methods of communication. The internet makes it much easier to inform prospective employers of one’s accomplishments, as a simple URL can now communicate what pages of printouts communicated before. In particular the new ICT related job positions, many of which did not exist yet in the time of AP’s data, require certain skills and characteristics that are particularly easy to communicate directly, now that we have electronic job applications. This does not mean that higher education is irrelevant to those professions, but it does mean that it is likely that we will not see ability signaling in those occupations. This fits well within the original signaling model by Spence: employees will not invest several tens of thousands of dollars in a signal if there are cheaper methods to signal their abilities in a more credible way. An example of such a cheap, credible signal is the LinkedIn network of recommendations, where people can inform the outside world of the skills of their associates, colleagues and friends. It seems as though in the modern labor market, these are the credible signals employers look for in their hiring process.

16 diminished educational quality difference between private and public education in terms of minimum skills learned. Open access to teaching methods, student performance and similar information have pushed both private and public educational institutions to uphold a certain standard of educational quality and skill guarantee upon graduation. This implies that students will look more at personal preferences and interests and less at a job-specific skill requirement when choosing a higher education. In turn, this makes the educational history less important in the job application process, as employers will expect all graduates, both public and private, to have at least the minimum skills required. A reduced exclusivity of higher education might lead to students trying to signal their abilities in other ways, such as extracurricular abilities and internships. The Chinese higher education case is an example of such a development, where so many students graduate each year that the degree itself holds no distinguishing value, and students flock abroad to use international experience as a signal to employers.

Another curious finding is the negative effect of private education on wages in early years for high levels of education. One possible explanation for this can be found in the motives behind choosing for private education. If people choose for private universities not because they are better educationally but because they are more prestigious, then there might be a negative selection bias in private education. It is commonly believed that private universities are more ‘elitist,’ where attendants place more value on secondary benefits from education, such as family status from attending the same prestigious university as your parents and individual prestige from having attended a high-ranking university. Yet in terms of productivity per invested tuition-dollar, public universities might overall generate students that are interested in getting a high-paying job rather than a prestigious network.

Using the data available in the NLSY97 surveys, I perform a quick regression to test this theory. Table 2 to the right shows regression results relating private educational attainment (𝑝𝑖) on skills (𝑧𝑖), household

income at the first interview round, highest educational grade completed by the father at the first interview round and real wage, as a simplified indicator of expected income gains. What table 3 shows reinforces the explanation put forward in the previous paragraph. Aside from skills and wage prospects, children from fathers who are highly educated are more inclined to go to private universities. If we can use father’s education as an

instrument for a child’s social class, then we can say that private educational choices are more determined by personal and relational preferences than a person’s foreseen career. A lack of professional ambition in educational choice would also mean that a credible signal would be non-existent, as private-university graduates would not have a selection bias in being more

17 productive, but rather in being more interested in secondary benefits from attending private universities.

This however leads to a completely different discussion that is not necessarily within the scope of this article. What these findings do indicate is that we need to back to the fundamental application of the signaling model within the modern economy. The assumption that I started my hypothesis with, namely that education serves a dual role of providing skills and sorting skilled people might not hold any longer in our modern economy. From the results of this article, it seems reasonable to hypothesize that skilled, highly-educated are still sorted into jobs with a faster wage growth path, but aside from opening doors to long-term promotions, education provides much less direct benefit than before.

CONCLUSION

The overall picture is then that using the current data, I fail to confirm the hypotheses that I put forward earlier in this article. The first hypothesis, that there is a signaling effect from private education, is rejected on the basis that I do not find employer learning over time and I do not find a reducing wage premium from private education with experience. Furthermore, there is no wage premium on higher education, nor on private education, which indicates that there is no ability signal nor a permanent productivity premium, but rather a productivity path that reveals itself with time. The hypothesis that there is a general signaling effect from educational attainment is also one I reject, again on the basis that there is no indication of employer learning of employee’s skills with experience. The prerequisite that the effect of schooling on wages should decrease with experience is met weakly in the analysis, but for this to translate into a signal, we require there to be an initial wage premium on education which is not present. Finally, I reject the hypothesis that education has a permanent positive on wages through a productivity bonus from schooling. Schooling without working experience does not lead to an increase in wages, and thus productivity, in my data.

The conclusion regarding the research question is then that there is no evidence that there is an ability signal premium in the wage build-up of private education graduates. It is however unclear whether this is due to inconsistency in the data, which replicated AP analysis hints at, or whether there is a structural effect within the labor market and the educational choice culture that is causing these findings. The implications of this conclusion for policy makers is that if they wish to improve employee productivity and reallocation in particular, then promoting private education is not a viable policy. This paper suggests that people primarily choose private education on the basis of their personal values and interests, and not so much because of their higher expected productivity upon graduation.

LIMITATIONS AND FUTURE RESEARCH

18 from the same timeframe, then we can say with more certainty that this is due to recent developments in the labor market and not due to problems with the model used in this paper. Additionally, in light of the findings of this paper, going back to the basics of the employer learning model might be important in uncovering the fundamental mechanics of the labor market in recent times, and whether or not these have changed in recent years.

19

REFERENCES

Altonji, J. G., & Perriet, C. R. (2001). Employer Learning and Statistical Discrimination. The

Quarterly Journal of Economics, 116(1), 313-350.

Arcidicono, P., Bayer, P., & Hizmo, A. (2010, October). Beyond Signaling and Human Capital: Education and the Revelation of Ability. American Economic Journal: Applied

Economics, 2(4), 76-104.

Bagger, J., Fontaine, F., Postel-Vinay, F., & Robin, J.-M. (2014). Tenure, Experience, Human Capital and Wages; A Tractable Equilibrium Search Model of Wage Dynamics.

American Economic Review, 104(6), 1551-96.

Bhandari, R., & Blumenthal, P. (2011). International students and global mobility in higher

education: National trends and new directions. New York: Palgrave Macmillan.

Bräuninger, M., & Vidal, J.-P. (2000, September). Private versus public financing of education and endogenous growth. Journal of Population Economics, 13(3), 387-401.

Clarke, M. (2007). The Impact of Higher Education Rankings on Student Access, Choice, and Opportunity. Higher Education in Europe, 32(1), 59-70.

Fang, H. (2006, November). Disentangling the college wage premium: estimating a model with endogenous education choices. International Economic Review, 47(4), 1151-1185. Farber, H. S., & Gibbons, R. (1996). Learning and Wage Dynamics. The Quarterly Journal of

Economics, 111(4), 1007-1047.

Heyneman, S. p., & Stern, J. M. (2014, March). Low cost private schools for the poor: What public policy is appropriate? International Journal of Educational Development, 35, 3-15.

Moro, A., & Norman, P. (2004). A general equilibrium model of statistical discrimination.

Journal of Economic Theory, 114(1), 1-30.

Seneca, J. J., & Taussig, M. K. (1987). Educational Quality, Access, and Tuition Policy at State Universities. The Journal of Higher Education, 1(58), 25-37.

Spence, M. (1973, August). Job Market Signaling. The Quarterly Journal of Economics, 87(3), 355-374.

Thomas, S. L. (2000). Deferred costs and economic returns to college major, quality and performance. Research in higher education, 41(3), 281-313.

Thomas, S. L., & Zhang, L. (2005). Post-baccalaureate wage growth within 4 years of graduation: the effects of college quality and college major. Research in Higher

20 van Ark, B., O'Mahony, M., & Timmer, M. P. (2008). The Productivity Gap between Europe and the United States: Trends and Causes. Journal of Economic Perspectives, 22(1), 25-44.

Vandenberghe, V., & Robin, S. R. (2004, August). Evaluating the effectiveness of private education across countries: a comparison of methods. Labour Economics, 11(4), 487-506.

Weiss, A. (1995). Human capital vs. signalling explanations of wages. The Journal of Economic

21

APPENDIX

APPENDIX 1: SUMMARY STATISTICS

MEAN ST.DEV MIN MAX N

𝒘_𝒊 10.488 8.0271 5 99.99 57286

𝒔_𝒊 12.538 8.4628 0 20 93962

𝒛_𝒊 42622 28652 0 100000 75014

𝒑𝒊 2.9710 1.7480 1 8 12384

𝒙𝒊 245.71 183.07 0 1113 97002

Note that the minimum values for wage 𝑤 and private education grades 𝑝 are not zero because those are treated as missing values instead.

APPENDIX 2: REGRESSION RESULTS

(1) (2) (3) (4) 𝒔_𝒊 -0.0437** -0.0405** 0.0891 0.0882 (-2.86) (-2.64) (0.92) (0.91) 𝒛_𝒊 -0.0203 -0.0313* -0.123* -0.115* (-1.48) (-2.16) (-2.53) (-2.24) 𝒑_𝒊 -0.307*** -0.308*** (-3.77) (-3.78) 𝒙_𝒊 0.359*** 0.350*** 0.514*** 0.500*** (9.70) (9.43) (4.18) (3.97) 𝒔_𝒊∗ 𝒙_𝒊 -0.106** -0.0610 -0.278* -0.284* (-2.68) (-1.39) (-2.02) (-2.06) 𝒛_𝒊∗ 𝒙_𝒊 -0.0511* 0.0290 (-2.32) (0.52) 𝒑_𝒊∗ 𝒙_𝒊 0.0505 0.0439 (0.73) (0.62) 𝒕 -0.300*** -0.291*** -0.292 -0.279 (-7.42) (-7.16) (-1.49) (-1.41) 𝒔_𝒊∗ 𝒕 0.343*** 0.301*** 0.0579 0.0644 (7.37) (6.04) (0.26) (0.29) 𝒛_𝒊∗ 𝒕 0.134*** 0.187*** 0.228*** 0.196** (10.44) (7.09) (5.28) (2.62) 𝒑_𝒊∗ 𝒕 0.425*** 0.432*** (3.71) (3.75) 𝒖_𝒊 0.0452*** 0.0453*** 0.0838*** 0.0833** (5.12) (5.13) (3.30) (3.28) _CONS -0.279 -0.280 2.870*** 2.865*** (-1.50) (-1.51) (3.91) (3.90)

T statistics in parentheses. * p<0.05, ** p<0.01, *** p<0.001. Coefficients for 99 occupational code dummy variables have been omitted from these results. [] ∗ 𝑥𝑖 and [] ∗ 𝑡 indicate interaction terms for experience and time, respectively. 𝑢𝑖