Returns to Education in UK - Essay Example

Returns to Education in UK Introduction Wages differentials and human capital framework is important in microeconomics decision and policy making. Adam Smith theory of wages and profits states that labour wages do vary with the labour skills, and ease or hardship of the work done (Smith, Edwin & Lerner) yet, human capital, as an investment, requires a cost of earnings forgone and pays a return  over the lifetime whether monetary or non-monetary. The analysis of returns to education has been one of the most studied relationships in economics literature. Returns to education studies are important to individuals and nations. Individuals need to understand whether education has a private economic benefit to them while nations need to understand whether investing in educational facilities will offer the same benefits to individuals thus motivate them to get more education. . The purpose of this study is to examine the returns to education in the UK. The next section presents the literature review where studies on returns to education are reviewed. Section 3 is the methodology where the research paradigm is presented. The section will also provide justification for choice of methodology, limitations of the design, validity, reliability and ethical issues considered in the study. Section 4 is the findings where results of the analysis are shown and interpreted. Section 5 is the conclusion where the summary of findings, limitations of the study, suggestions for policy and practice, and suggestions for further research is made. . 2 Literature Review Research to estimate returns to human capital have focussed on the Mincer function which decomposes the periodic earning into a linear functional form. The log of earnings (dependent variable) is a function of schooling, years of experience in work and other factors representing heterogeneity amongst the cross-section sample of people observed such as race or gender (Harmon et al). The model has some limitation namely biasedness in the estimation of returns to education and endogeneity in the schooling variables for two main reasons. Firstly, unobserved abilities to earn higher wages may also correlate with learning, so some individuals stay in education longer and also earn higher wages in their occupation, meaning both schooling and wages are positively correlated when actually both are a function of the unobserved variable of ability. Secondly, existence of a systematic relationship between any of the independent variables such as schooling and the error term in an Ordinary Least Squares regression will lead to bias in the estimates. In this case, effects of ability/heterogeneity must be random in the sample and normally distributed to avoid positive correlation (between schooling and ability) leading to a higher than estimated return to education. Card explored the relationship between education and earnings, and explicitly analysed the heterogeneity between schooling of twins in contrast to their earnings. The assumption in the study was that twins would have the same ability and other external influences so that differences in wages could be more accurately associated with differences in education. Card used a pooled sample of 198,075 men and women aged from 16 – 66 during the years 1994 to 1996. The study targeted the twins having 10, 12 and 16 years of schooling, and the earnings differences between them. An Ordinary Least Squares regression analysis was used to inspect the human capital return with hourly, weekly and annual earnings as alternative variables. The study findings explored an interesting impact of an instrumental factor family background (Altonji) on the schooling and earning of their children which had 30% variation in the earnings, similarly college education differences and location near college or university had some significant influence over schooling and hence earnings. Birth season was only an insignificant instrumental factor (Card). The Ordinary Least Squares estimator results by Card suggested a difference of around 10% in the earnings of twins that could be explained by educational differences. Furthermore the earnings of certain subgroups that had external structural influence on their schooling were higher than average individuals as a whole. Dearden et al applied Ordinary Least Squares estimation to returns to academic as well as vocational qualification in Britain. This study used a longitudinal sample from the National Child Development Study (NCDS), 1991 and Labour Force Survey (LFS) of 1998. Their study examines the varying aspects of gender, ability, qualification and family background too. This longitudinal study develops a framework where NCDS results were compared with LFS statistics while after some selection criteria a final sample size of 6867 individuals (3007 Male & 3860 Female) was processed through estimation. Estimation results identified that males with O-Levels/ GCSE earn 13 to 23% premium to their qualification while for A-levels they earn 18% and for a degree the premium was 11 to 32%. In contrast to men, women with O-levels/ GCSE earn 11 to 21% premium and with A-levels it adds to 20% and for degree women return 30% premium to the qualifications in a year. The study confirmed that along with academic qualifications, if individuals acquire vocational qualifications their premium may rise by 40% but vocational qualifications alone have reported fewer premiums (Dearden, Mcintosh and Myck). This research identified a difference between male and female premiums to their qualifications where female earn less to their schooling. Further continued study analyses to the same domain were presented in McIntosh. He used the sample data of Labour Force Survey during the period 1996 to 2002. His study also applied OLS estimation after reducing biases and error. The sample size stood at 15,000 individuals. His study aimed to focus on the changing patterns of return to schooling. The results revealed an identical premium return to both males and females. Yet the variance of qualification confirmed heterogeneous returns among O-levels, A-levels and degree level labour force (McIntosh). The study found that there were steady returns to academic qualifications in general. In other words, while the demand for higher level of academic qualifications rose, the supply of skilled labour also rose thus making the returns to education remain steady. The study estimated Ordinary Least Squares regression equations for each of the years separately by converting quarterly data into annual data. The researcher also estimated the results for those who were full-time employees only. Education was measured using the variable ‘all qualifications held’ as opposed to ‘highest qualification held’ as the study noted that the later measure had some limitations. These were included in logarithmic wage equations controlling for ethnicity, region, and quadratic in age. Human capital investment is a private decision which affects the return to schooling; so, every additional qualification adds to the return (Dickson and Sarah). Harmon et al., conducted microeconomic empirical investigation on the returns to schooling. They concluded that every investment reflects the size of returns. Harmon used a comparative approach to analyse cross sectional returns on education and furthermore, applied Ordinary Least Squares method using a dataset acquired from International Social Survey Program (ISSP). The results revealed relative differences among the twenty five countries mostly from European Union (EU) such as Ireland, United Kingdom, Italy, Germany, Greece, Norway and Switzerland based on their microeconomic wage circumstances. The overall average return to schooling through Ordinary Least Squares statistics was concluded around 6 to 11% international. A number of studies have investigated regional differences using a dummy variable to control for the effect of education on wages (see for example McIntosh). In a study on regional returns to education in India, Chamarbagwala found evidence that higher regional returns to education increased the likelihood that children go to school while reducing the likelihood that they work. This study is important as it included regional dummies as independent variables in a model and tested for its effects on a dependent variable. While it is clear that region has been studied in a number of studies on returns to education, most of them do not report the exact relationship between the dependent variable and region hence it is very difficult to show for certainty whether its effect has been positive or negative but it has been significant. For instance, the current British Household Panel Survey dataset has region as a main variable and it is assumed that those working from London are more likely to earn higher than those working elsewhere in the United Kingdom. The hypothesis therefore tests whether this is true. It is expected that longer years of education for those working in London will lead to higher hourly wages. I therefore test the hypothesis that region has a significant impact on the hourly wages in the United Kingdom. The studies reviewed have also shown the influence on race on the dependent variable – wages. Just as in region, race has been included as a control variable in most of the studies on returns to education but its effects have not been directly reported in most of them. Most of the studies have just cited that its effect was significant but the direction of the effect on wages has not been expressly provided. In this study, race has been defined as a dummy variable of 1 for Whites and 0 for the rest of other races. It is expected that White workers will have higher hourly wages than other races and therefore a positive effect is expected. I therefore test the hypothesis that race has a significant effect on hourly wages in the United Kingdom at a confidence level of 0.05. The review also shows that experience of workers has been used as a variable in almost all of the studies on returns to education. The results have shown mixed findings as some have noted positive effects, others negative, while others no significant results. The variable has also been operationalized differently in studies which could explain the mixed findings. In the present study, I examine the effect of experience on hourly wages using two variables – experience and experience2. Experience is defined as the age of a person less 12 years of compulsory schooling. I assume that the difference is the experience of a person since that is the minimum requirement for a job in the UK. I expect to find positive effects of both experience and experience2 on hourly wages as more experience should lead to more hourly wage rates. I therefore test the hypothesis that experience has a significant influence on hourly wages. 3 Methodology This study explains whether schooling offer private economic returns. The study is therefore a quantitative research as it uses the BHPS data collected. This has been chosen because the intention here is to analyse and quantify a relationship between variables and therefore quantitative research paradigm is suited for the present study. Using secondary data which was collected in another survey – British Household Panel Survey, I expect that the ethical considerations were made when the data was being collected and therefore I do not have to worry consider these for the estimation itself. Further, I am relying on this data because of its high validity and reliability as it was from a national survey in the UK. The sample used was about 5,000 households and covered over 10,000 individuals selected randomly from the population. The survey ran from 1991 – 2008 Concerns of the scholars that bias or endogeneity existed in estimating returns to schooling did not prevent them from carrying out their regression analyses. One approach avoiding endogeneity is to use instrumental variables. These are variables that are correlated with ability but not schooling for example. By including these, the Mincer regression model is extended to include some proxy measure for ability. The typical Mincer regression model for estimating returns to education for individuals is (Harmon) Ln(wi)=β0+β1Educi+β2expi+β3exp2i+Xβ+εi (equation 1) Where: i represents each person in a cross section such as the BHP survey data; Ln(wi) = the logarithm of wage for person i; Educ = is the variable measuring years of schooling or higher education; exp = years of experience; X is a vector of other characteristics describing heterogeneity in the sample that might be correlated with earnings such as gender and marital status; and ε is the error term in the regression equation. The hypothesis is β1 is positive and significant which represents the returns to education. In this study, we test the relationship between education (measured as years of schooling) and returns (measured as logarithm of hourly wage) and controlling for sex, race, and region. 4 Findings This chapter presents the results of the study. Figure 1 shows the distribution of returns (measures as hourly wage). As shown, the hourly wage distribution was skewed to the left. A logarithmic distribution of hourly wage in Figure 2 shows that the distribution is more normal. Figure 3 shows the distribution of years of schooling which is not normal. These variables could be used in the OLS regression which assumes normality of distribution in the error term. Figure 1: Distribution of hourly wage Figure 2: Distribution of logarithm of hourly wage Figure 3: Distribution of Years of Schooling Table 1 shows the descriptive results on the variables used in the study in terms of mean, median, min, max, standard deviation, skewness, and kurtosis. Further, the normality of distributions is also tested using Jarque-Bera tests. Sum of variables and the number of observations for each variable is also shown. Age as used as direct data of years of work experience was not present in the BHPS survey. Age may overestimate experience if individuals spent some years unemployed. Table 1 shows that the mean and median age was 43 years with a standard deviation of 10 years. Age of the participants ranged from 25 – 77 years. Results also show that hourly wage ranged from 0.54 – 72.17 with a mean of 10.14 and a standard deviation of 6.18. The median wage per hour was 8.53. It was revealed that years of schooling ranged from 5 – 22 with a mean of 12.72 and a standard deviation of 2.58. The median years of schooling was 12 years. The Jarque-Bera test of normality shows that age, wagehr and years of schooling were not normally distributed as the p-values were less than 5%. Table 1: Descriptive Summary Results AGE WAGEHR YEARSOFSCHOOLING  Mean  43.22620  10.13866  12.72023  Median  43.00000  8.534993  12.00000  Maximum  77.00000  72.17090  22.00000  Minimum  25.00000  0.537037  5.000000  Std. Dev.  10.22252  6.179559  2.582733  Skewness  0.263237  2.463214  0.748368  Kurtosis  2.371390  13.93856  2.828995  Jarque-Bera  82.10773  17576.45  277.1583  Probability  0.000000  0.000000  0.000000  Sum  126696.0  29716.43  37283.00  Sum Sq. Dev.  306185.0  111887.8  19544.59  Observations  2931  2931  2931 A covariance analysis results in Table 2 show the correlation matrix for all the variables in the study, the t-statistic, probability, and observations. Table 2: Covariance Analysis Covariance Analysis: Ordinary Date: 12/04/13 Time: 15:21 Sample: 1 5323 Included observations: 5323 Pairwise samples (pairwise missing deletion) Correlation Probability Observations LOG(WAGEHR)  AGE  FEMALE  LONDON  WHITE  YEARSOFSCHOOLING  LOG(WAGEHR)  1.000000 -----  2975 AGE  -0.089124 1.000000 0.0000 -----  2975 5323 FEMALE  0.296404 -0.025137 1.000000 0.0000 0.0667 -----  2975 5323 5323 LONDON  -0.008367 -0.003489 0.003005 1.000000 0.6484 0.7992 0.8266 -----  2972 5318 5318 5318 WHITE  -0.017033 0.085524 -0.012214 0.039559 1.000000 0.3532 0.0000 0.3730 0.0039 -----  2973 5321 5321 5316 5321 YEARSOFSCHOOLING  0.410493 -0.385261 0.029142 -0.032524 -0.118078 1.000000 0.0000 0.0000 0.0346 0.0184 0.0000 -----  2935 5260 5260 5255 5259 5260 The results in Table 2 show that hourly wage was positively and significantly correlated with Years of Schooling and Sex (female = 1). Hourly wage was negatively and significantly correlated with age, region (London=1) and race (white=1). The correlations between the predictors in the model were low reducing the chance of serial correlations that would lead to spurious estimations in the OLS regression. Equation 1 was estimated using normal OLS and a heteroskedasticity test was conducted using White test the test value (Obs x R2 was 25.24 which was distributed as a Chi-squared with 20 degrees of freedom. This was below the 5% critical Chi-squared value of 31.4 so heteroskedasticity was rejected (see Table 3) and the normal OLS results reported in Table 4. Residual plots in Figure 4 also suggest homoscedastic errors. Table 3: White Heteroskedasticity Test Heteroskedasticity Test: White F-statistic 1.263903     Prob. F(20,2903) 0.1922 Obs*R-squared 25.24114     Prob. Chi-Square(20) 0.1924 Scaled explained SS 46.68992     Prob. Chi-Square(20) 0.0006 Test Equation: Dependent Variable: RESID^2 Method: Least Squares Date: 12/04/13 Time: 15:52 Sample: 13 5323 Included observations: 2924 Collinear test regressors dropped from specification Variable Coefficient Std. Error t-Statistic Prob.   C 0.499915 0.281385 1.776621 0.0757 YEARSOFSCHOOLING^2 0.000235 0.000774 0.303521 0.7615 YEARSOFSCHOOLING*FEMALE 0.003603 0.005936 0.606861 0.5440 YEARSOFSCHOOLING*AGE(-12) -0.000275 0.000904 -0.304753 0.7606 YEARSOFSCHOOLING*(AGE(-12)^2) 2.06E-06 8.30E-06 0.248001 0.8042 YEARSOFSCHOOLING*LONDON -0.011671 0.015025 -0.776782 0.4374 YEARSOFSCHOOLING*WHITE 0.010407 0.012630 0.823942 0.4100 FEMALE^2 -0.035419 0.201738 -0.175572 0.8606 FEMALE*AGE(-12) -0.003321 0.006254 -0.530985 0.5955 FEMALE*(AGE(-12)^2) 3.77E-05 5.65E-05 0.668473 0.5039 FEMALE*LONDON 0.110323 0.071373 1.545725 0.1223 FEMALE*WHITE 0.039653 0.086777 0.456953 0.6477 AGE(-12)^2 -0.000148 0.000428 -0.345703 0.7296 AGE(-12)*(AGE(-12)^2) 3.00E-06 7.71E-06 0.389764 0.6967 AGE(-12)*LONDON 0.019283 0.015385 1.253319 0.2102 AGE(-12)*WHITE 0.004500 0.017413 0.258456 0.7961 (AGE(-12)^2)^2 -1.83E-08 3.63E-08 -0.504103 0.6142 (AGE(-12)^2)*LONDON -0.000152 0.000138 -1.099336 0.2717 (AGE(-12)^2)*WHITE -2.85E-05 0.000155 -0.183576 0.8544 LONDON^2 -0.455764 0.440363 -1.034974 0.3008 WHITE^2 -0.393480 0.490218 -0.802664 0.4222 R-squared 0.008632     Mean dependent var 0.211945 Adjusted R-squared 0.001802     S.D. dependent var 0.408565 S.E. of regression 0.408197     Akaike info criterion 1.053023 Sum squared resid 483.7120     Schwarz criterion 1.095976 Log likelihood -1518.519     Hannan-Quinn criter. 1.068493 F-statistic 1.263903     Durbin-Watson stat 1.924296 Prob(F-statistic) 0.192160 In Table 4, the results show that years of schooling positively predicted log(hourly wage), p < 0.05 showing that education had positive returns. From the coefficient of years of schooling, it can be observed that a year increase in years of schooling leads to an 8.4% rise in hourly wage, holding all other factors constant. Table 4 shows that sex (female =1) has a positive and significant effect on hourly wages, p < 0.05. This means that females in the in UK appear to earn 30.5% higher wages holding other factors constant. Results also show that experience (as measured by age minus 12 years of basic schooling) was positively but insignificantly correlated with hourly wages, p > 0.05. Age squared (or experience squared) was negatively correlated with hourly wages but the relationship was weak, p < .10. These results on experience show that hourly wage rate in UK is not significantly influenced by the level of individual experience. Table 4: Regression Results Dependent Variable: LOG(WAGEHR) Method: Least Squares Date: 12/10/13 Time: 23:32 Sample (adjusted): 13 5323 Included observations: 2924 after adjustments Variable Coefficient Std. Error t-Statistic Prob.   C 0.710347 0.112345 6.322899 0.0000 YEARSOFSCHOOLING 0.083802 0.003317 25.26489 0.0000 FEMALE 0.304622 0.016962 17.95942 0.0000 AGE(-12) 0.005569 0.003482 1.599221 0.1099 AGE(-12)^2 -5.33E-05 3.14E-05 -1.695929 0.0900 LONDON -0.010331 0.039824 -0.259409 0.7953 WHITE 0.123498 0.047149 2.619293 0.0089 R-squared 0.252075     Mean dependent var 2.170926 Adjusted R-squared 0.250536     S.D. dependent var 0.528811 S.E. of regression 0.457800     Akaike info criterion 1.277624 Sum squared resid 611.3484     Schwarz criterion 1.291942 Log likelihood -1860.887     Hannan-Quinn criter. 1.282781 F-statistic 163.8536     Durbin-Watson stat 1.801133 Prob(F-statistic) 0.000000 Table 4 also shows that region (as measured by a dummy variable of 1 for London and 0 otherwise) did not significantly predict hourly wage in UK, p > 0.05. This means that those working in London do not earn significantly higher hourly wages than those outside London. Working in London does not therefore influence the hourly wages of workers in UK. Lastly, the results show that race (as defined as a dummy variable of 1 for white and 0 otherwise) positively influenced hourly wages, p < 0.05. This means that white workers were more likely to earn higher wages than non-white workers. Thus, a one year increasing in schooling leads to a 12.3% rise in hourly wages for white workers in Britain. Table 4 show that R2 is 0.25 suggesting the model explains only 25% of the variance in log(wagehr). Thus, most of the variation in wages remains unexplained. Table 5 shows that the F statistic was 3038.01 and significant at below 1% level. The regression is therefore significant in explaining log( wagehr) and model is robust. Table 5: F-Test Wald Test: Equation: EQ03 Test Statistic Value df Probability F-statistic  3038.012 (6, 2917)  0.0000 Chi-square  18228.07  6  0.0000 Null Hypothesis: C(1)=C(2)=C(3)=C(4)=C(5)=C(6)=C(7) Null Hypothesis Summary: Normalized Restriction (= 0) Value Std. Err. C(1) - C(7)  0.586849  0.140522 C(2) - C(7) -0.039696  0.046773 C(3) - C(7)  0.181124  0.049942 C(4) - C(7) -0.117929  0.047271 C(5) - C(7) -0.123551  0.047150 C(6) - C(7) -0.133829  0.062943 Restrictions are linear in coefficients. Figure 4: Residual Plots 5 Conclusion This study has analysed returns to education in the UK. This was done using the BHPS data from 1991-2008 by using an OLS regression model adapted from previous studies. The dependent variable was logarithm of hourly wage while the predictor variable tested was years of schooling, controlling for other factors. The control variables were age, sex, race and region. Diagnostic tests was conducted to check for heteroskedascity and also the presence of serial correlations and both were found not to affect the model hence the regression was run. The analysis shows that the number of years in schooling had a positive and significant effect on the logarithm of hourly wage (about 8% per year of schooling) which suggests that there were positive returns to education in the sample. This is consistent with a number of studies that have analysed the returns to education. From the R-squared results, the model accounted for 24% of the variance in hourly wages. Thus, the regression did not explain most of the variance in hourly wages but points to the fact that the number of years of schooling is a good predictor of returns in education. However, we cannot rule out endogeneity (race results) and given low overall significance of regression (R2 = 25%. These results must be treated with some caution. Comparing with the results of Harmon who found that returns to education are between 6-11% across different countries the results this regression (8% returns to education) compares favourably and are consistent with other studies. This study contributes to the literature on returns to education by estimating how years of schooling are a significant contributor of economic returns to education as moderated by age (experience), sex and region. This is important for both theory and practice. Policy makers therefore can make investment decisions with the conviction that education pays off ultimately through private economic benefits. A major limitations in this study was the use of years of schooling to proxy education qualifications or credentials which may presents the “sheepskin effect” problem as explained by Card. Future studies should therefore employ other variables to proxy education as opposed to using the years of schooling. Further studies should also focus on whether education leads to better social returns more than private returns especially in developed nations. Read More