Earnings And The Value Of A Canadian University Degree James McIntosh Economics Department Concordia University 1455 De Maisonneuve Blvd. W. Montreal Quebec, H3G 1M8, Canada. Danish National Institute of Social Research Herluf Trolles Gade 11 DK-1052 Copenhagen K, Denmark February 16, 2009 E-mail Addresses and Telephone Numbers: jamesm@vax2.concordia.ca. 514 848 2424 3910. Keywords: Canada, Earnings functions, Mincer equations, Ability, and Unobservable Heterogeneity. JEL Classification Numbers: J24, J30. Abstract Using methods which deal with unobserved worker characteristics I am able to identify three distinct earnings groups of fully employed Canadian males who were included in the 1973 and 1997 Survey of Consumer Finances. While mean real earnings exhibit an increase of 5.4% over this period my results show that the two survey years are rather different in terms of the composition of ability or skill groups. In 1997 the high earners, the superstars who had earned much more than the other two types in 1973, had disappeared and were replaced by a low income group. In 1997 this group, which made up 22.3% of the sample, was earning 22.8% less than average. The present value of earnings over the ages 25-60 also declined for all three groups so that in terms of remuneration the value of a university degree is less in 1997 than it was in 1973. Employment rates have also fallen for this group of elite workers so the value of a Canadian university degree has unambiguously declined over this period. 1 Introduction The university system in Canada has expanded dramatically in the last half of the twentieth century. More specifically, data from the Survey of Consumer Finances shows that in 1973 12.6% of males aged 25-34 had a university degree. By 1997 this figure had increased to 17.7%. Results from the 2002 Survey of Approaches to Educational Planning described in Shipley et al (2003, p. 8) 1 indicate that this trend will continue since a majority of parents, 61%, with children in secondary school wanted them to go to university. While most Canadians are enthusiastic about the success of Canadian higher education and appreciate the increased access to universities, a comprehensive and objective assessment of the real contributions of the expansion of university system has yet to be made. To contribute to our understanding of this issue I examine an aspect of university graduates’ labour market performance. The question that this research addresses is how the earnings of male university graduates have evolved as the number of university degree holders has increased over this twenty-four year period1 . Knowing what a university degree is worth is essential for both rational individual decision making and government policy making. For Canada much of the research on earnings has been comparative and has concentrated on earnings differentials across educational categories so this topic has not been fully investigated. There are some results, however, on how real earnings have changed over the last twenty-five years. For the period 1980-2000 Chung (2006) reports a slight increase in average real weekly earnings for degree holders in the 25-54 age group working full time but a 5% reduction in their full-time paid employment. It would, therefore, appear that recent male graduates have not suffered any decline in real earnings. However, as will be seen, this apparent absence of a significant downward trend in average real earnings over all employees masks a serious reduction in the earnings of many degree holders. Using methods which deal with unobserved worker characteristics I am able to identify three distinct groups of individuals who were included in the Survey of Consumer Finances for 1973 and 1997. These groups appear in different proportions in the two surveys with the emergence of a low income group in the 1997 survey. Comparisons across the two surveys reveal lower discounted present values for all three groups in 1997. Employment rates have also fallen for university graduates over this period. Thus, some of the problems that researchers in the economics and sociology of education have noted may be having an impact on the earnings of university graduates.2 . 1 This study only deals with males. Preliminary investigations suggested that females needed to be treated separately. For example, females experienced increases in both real weekly earnings and employmnet rates over this period! It also concentrates on the top end of the labour market, those workes who earn more than the minimum wage and have full time jobs. 2 Côté and Allahar (2007) in their monograph on Canadian universities write on page 54 “Being able to play the system to make your way through university may be good training for some jobs that do not require integrity, but it is not beneficial for jobs that require the ability to manage one’s time and motivate oneself to complete tasks honestly and responsibly. The attendance and effort of many university students would get them fired if they were in the labour force; in many universities sporadic attendance and minimal effort often have little or no consequences in grades. ..... At the same time we all know that many students are slipping through our courses with minimal learning and skills mastery, ....The wrong signals are being sent to employers about the quality of some of our university graduates, and policy makers are misled about the value added by the surplus of university credentials. Put another way, we may not be producing the level and quality of human capital that employers and governments think we are.” 2 In this study earnings functions are estimated using data from the two surveys mentioned above. The dependent variable is the natural logarithm of total annual income from all sources. Only those males earning more than ten thousand real (2002) dollars per year and worked a full year were included in each sample. The explanatory variables included were an age variable, type of residential area, province, and mother tongue. The data is described in section 4. The rest of the paper is organized in the following way. Section 2 reviews the relevant economic theory and Canadian empirical work. The data is described in section 3. Section 4 outlines the statistical models employed in the analysis and the paper ends with a discussion of the results. 2 A Brief Review of the Theoretical and Canadian Empirical Literature How one views earnings functions depends on the assumptions made about educational decision making. There are several issues in this area that need to be examined. In earnings functions, like Mincer equations for example, are the schooling and experience variables endogenous or exogenous? The simplest way to examine these issues is to formulate a reasonable model of individual educational decision making. Here reasonableness means that the model be forward looking and sequential, taking into account the fact that current decisions have future consequences which are uncertain. To be specific, some notation and assumptions are required. Let St be the number of years of school that an individual has acquired by the beginning of year t, and dt be a decision variable that takes the value 1 if the individual decides to do an additional year of schooling and 0 otherwise. The cost of staying in school in year t is h(St , a). Following in the tradition of Spence (1973), education costs vary inversely with ability, a and I assume that costs increase with the amount of schooling. Individuals in school earn no income and are assumed to make education decisions which maximize the present value of discounted expected earnings. They have an economic life span of T (t), T 0 < 0, years. In this highly stylized model anyone who decides not to continue in school in year t immediately starts working and earns y(St , 0, a, ut ) in year t and works full time until t + T (t) − St and is never unemployed. The zero in y() refers to the initial level of experience. ut is a random and is known before the year t decision is taken. In keeping with the Spence view, those with more ability will earn more so that y() is increasing in ability. To keep the model relatively simple it is assumed that only complete years are considered, everyone starts school at age zero, those who leave school can never return, and all additions to human capital after the cessation of school are determined exogenously. Letting Ωt be the information set at the beginning of year t the value function 3 for the Bellman equation given by V (St , ut ) = M ax [dt {−h(St , a) + δE[V (St + dt , ut+1 )|Ωt , dt = 1]} dt ∈{0,1} T (t)−St + (1 − dt ){ Σ j=0 δ j E[y(St , j, a, ut+j )|Ωt , dt = 0]}] (1) where δ = 1/(1 + r) and r is the individual’s discount rate. The first term in the value function is the value of continuing one more year in school and the second term gives the present expected discounted value of earnings if the individual ceases school and enters the work force at the beginning of year t. The decision to continue in school in year t will be determined by comparing the two parts of V (St , ut ). If the first part is larger than the second then the individual will continue in school leaving the decision the leave school to year t + 1, otherwise he or she will leave school and start working. Clearly St+1 = St + dt dt = {0, 1} (2) and completed schooling is t∗ St∗ = Σ dt t=0 (3) and t∗ is the year in which the individual decided to finish school and start working. This model is quite well known. It is a version of the Mare (1980) grade transition model whose properties are discussed in great detail by Cameron and Heckman (1998) and Heckman et al (2006, p. 342-358). Dynamic models of earnings and together with various forms of human capital formation, educational, and occupational choice have also been estimated by a number of authors. A review of this literature may be found in Belzil (2006). The model has a number of features which inform researchers about the way earnings functions should be estimated. First, St∗ depends on {(us , Ωs ) s ≤ t∗ }, the potential earnings profiles, and, of course, ability. Hence, it is determined in a way which is much more complex than that suggested by static models, like those reviewed in Card (1999), for example, that are commonly employed to explain earnings. Secondly, it is clear from the structure of the model that the estimation of the parameters in the earnings function that St∗ and the amount of experience that the individual has accumulated should be treated as exogenous variables. They are determined prior to the time that earnings are observed3 . This feature of the model will allow earnings functions to estimated for separate educational categories, such as determined by the number completed years of schooling, without generating any selection problems. It is only when researchers are interested in the impact of the level education or the number of 3 Belzil (2006, p.7) makes this point in his 5th footnote where he says ‘Technically speaking, the term ”endogeneity” used in the empirical literature abuses the true meaning of endogeneity. In cross-section data wages are usually measured much beyond the time when schooling is completed. Because returning to school is rarely observed, schooling may be therefore be viewed as predetermined. It is only the correlation between schooling and unobserved skills that is problematic’ 4 completed years of schooling on earnings that the traditional unobserved ability bias problem needs to be addressed. With these theoretical perspectives in mind I now turn to a brief summary of the literature that examines the earnings of Canadians. Most of the Canadian research focuses on the size of the ‘wage premium’ paid to university graduates. This issue is examined in by Bar-Or et al (1995), Burbidge et al (2002), Morissette et al (2004) and Boothby and Drewes (2006). For the period 1971-1991 the first authors find a ‘U ’ shaped university premium for males which has a larger value at the beginning of the period. For females no clear results are obtained. These results are largely based on the data using graphical procedures although some of their results are obtained by regression methods. The next two papers fit Mincer type earnings functions and find a more or less constant wage premium over the period 1980-2000. Both of these papers consider various subpopulations like age groups, gender, or field or occupational category. None of these papers consider the problems that can arise when there are unobservable characteristics that affect earnings. In an important paper Green and Riddell (2003) fit Mincer equations augmented with some family background variables and a literacy test score for the respondent. They find, that as a measure of cognitive skills, it has a significant effect on earnings and this confirms the results that Osberg (2000) found. It also demonstrates the importance of using procedures which deal with unobserved ability. Unfortunately, as was shown in McIntosh and Munk (2007) including test scores may not be sufficient to solve all of the problems that arise from the presence of unobservable factors. 3 Data The data used in this paper comes from the Survey of Consumer finances for the years 1973 and 1997. 1997 was the last year that Statistics Canada administered this survey. It was started in 1971 and was initially run every two years after 1980 it was run every year. 1973 was the first survey that had large enough sample sizes for the analysis carried out here. Unfortunately, the surveys that replaced the Survey of Consumer Finances, The Survey of Labour Income Dynamics and The General Social Survey, do not collect the data in the same way making it problematic to make comparisons between the 1973 results and for years later than 1997.4 The earnings measure is total income from all sources before taxes. Real income was obtained by dividing this by the consumer price index (2002 = 100). For the sample only males earning ten thousand 1997 dollars are included. Some sample selection is required since some earnings are negative but the number here is small being less than 1%. To analyze the most productive part of the 4 There is data from the SLID for 1997 but it gives mean real earnings which are 25% higher than the SCF earnings data for the same year. It is, therefore, difficult to distinguish changes that arise as consequences of labour market behaviour from those that are due to differences in survey design. 5 labour force I chose only those who were employed all year and had significant earnings. The lower limit of ten thousand dollars was chosen since this is a good approximation of the earnings of an individual who earns the minimum wage. The selection problems that arise because of the lower bound on income are dealt with by using truncated distributions. Real earnings by age and year are displayed in Table 1 by 5 year intervals. The surveys also contain information on province of residence, size of the urban community in which the respondent resided and mother tongue. These variables are represented by sets of dummy variables and are used as regressors in all of the models. The employment rate is also shown. This is sometimes equal to 1 because of the small sample sizes in the 1973 survey. This is the proportion of university graduates who in full time employment. Mean earnings for each educational category are displayed in Table 1 as are the shares of each educational category. Only males aged 25-60 are included in the sample and the averages are taken over these ages. 4 Statistical Models Real earnings for individual i depend on his educational qualifications as well as his ability which is represented by the variable ai . A model that is much more general than the established Mincer equation is described by the equation ln(yi ) = Xi γ + α0 (ai ) + α1 (ai )zi + α2 (ai )zi2 + (4) where the variable zi is (Agei − 24) and the functions α0 (ai ), α1 (ai ), and α2 (ai )depend on the level of educational attainment. This variable, zi , plays the role of experience in the model. While there was no data on actual work experience or the age when the agent started full time work, all the agents that are being examined here have a university degree and reported as having worked for at least 48 weeks in the year prior to the interview. Consequently, zi will be highly correlated with actual work experience. The idea here is that ability and education interact differently at each level of educational attainment so that there is a version of equation (4) with its own specific set of {aj (ai ), j = 0, 1, 2} for each level of education. Furthermore, ability has an age or experience specific effect on real earnings. Turning now to the question of how the parameters in equation (4) should be estimated, the first point to note is that since the focus of interest is on the value of a university degree only the earnings equation for university graduates will be estimated. As was pointed out in section 2 it is possible to estimate earnings functions for university graduates as a separate group without encountering any biases due to this selection process because, with respect earnings, the level of education is exogenous. The question that remains to be answered is how to deal with the unobservable ability variable, ai . One procedure that can be used to deal with its effects is to follow Heckman and Singer (1984) and assume that ai takes on a finite number of values {ai = a1 , a2 , ....aL } where Pr{ai = a , = 1, 2...L} = p and 6 P p = 1. This means that there are a finite number of ability types and their earnings functions have means µi = Xi γ + α0 + α1 zi + α2 zi2 (5) if individual i is of type . This leads to a model which is a generalization of the Heckman-Singer (1984) procedure and belongs the latent class models discussed by Wedel et al (1993) and Deb and Trivedi (1997). This means that a type worker has a specific response to his educational attainment and age which depends on his individual characteristics. Those with more ability can be expected get more out of their educational qualifications and experience. The term Xi γ represents other characteristics of the individual: province, mother tongue, and urbanization dummies. While it is not possible to assign a type to each individual the procedure assumes that individual i will be of type with probability p .These probabilities can then be estimated along with the type parameters by maximizing the sample likelihood function whose natural logarithm is ln(L) = N P i=1 ln[ L P p{ =1 φ(ln(yi ), µi , σ ) _ }] 1 − Φ(ln( y), µi , σ ) (6) where L and N are the number of types and the sample size, respectively and _ density functions, Φ(ln( y), µ , φ(ln(yi ), µi , σ ) are normal i σ ) is the cumulative _ normal distribution, and y is 10,000. The choice of the number of mixtures to apply is an empirical issue to be determined by criteria involving the value of the maximized likelihood together with the number of parameters. The appropriate model to be selected is determined by the data in Table 4. The first line contains the value of the maximized likelihood function using a single truncated distribution with no covariates except a constant term and a variance parameter. This serves as a baseline which can be used to compare other models and to construct a pseudo-R2 for each model. Additional mixtures were added until there was no significant increase in the penalized likelihood function or until one of the probabilities converged to zero. For both surveys 3 mixtures were all that could be estimated. 5 Results and Discussion Maximum likelihood estimates for the parameters of interest for two surveys are displayed in Tables 2 and 3 for each mixture distribution. Age profiles are significantly concave for all types in each survey and the parameters associated with z and z 2 are highly significant for all types in each survey. Types differ significantly by at least one estimated coefficient and there are significant differences in the estimates of the standard deviations across the three types in each survey. The estimates for the parameters associated with the province or the size of the locality in which the individual lived, the respondent’s mother tongue are not included since they are of no particular interest and their inclusion would make the tables reasons already mentioned. 7 In each survey there are three distinct ability types. It is clear from Table 4 that the increases in the ln-likelihood function justify the increase in the number of mixing distributions5 . For the data the process stopped at three for each survey because the probability of the fourth mixture converged to zero when a four mixture model was estimated or the procedure would not converge when four mixture models were being estimated. The reason for this is that the four mixture latent class model appears not to be identified. There may, in fact, be more than three ability types but the data is not rich enough to identify them. The ln-likelihood values in Table 4 also generate the psuedo-R2 statistics. The model explains much more of the variation in the data in 1973 than in 1997 but given the type of data that is being examined here the models do an acceptable job of explaining it. In Table 5 predicted average earnings profiles for all three types are displayed for university graduates for each of the two surveys. Because of the truncation in the distributions these are computed as _ yb = exp[µ + σ { φ(ln( y), µ ) _ }] 1 − Φ(ln( y), µ ) (7) Looking first at the 1973 data, of the three types, type 2 and type 3 are very similar. For this survey the increase in the ln-likelihood function that results when the third type is added is significant but quite small. Type 1, on the other hand, earns much more than the other two. It is for this reason that the individuals in this category are referred to as superstars and by age sixty earn two and a half times what the other groups earn. Superstars make up 6.8% of the sample. For the 1997 there are two groups which earn a little more than average earnings. There there are no earnings superstars; instead a low income group has emerged. They are much more numerous than the 1973 superstars and earn 22.8% less than average earnings. They make up 22.3% of the sample. This is a significant decline but it represents at best a lower bound on the extent of the devaluation of university education. The situation could actually be worse. This study focuses on only those who were lucky enough to have found employment for the full year. Some university graduates are either not employed for fifty-two weeks or make less than ten thousand dollars per year. In 1997 the 19.1% of all males aged 25-60 in the survey were excluded because they did not meet either the employment or the income constraint. This is a substantial increase over the corresponding rate of 10.1% for the 1973 data. Finally, a caveat is in order. The discussion about types has focused on a rather abstract notion which I have referred to as ability or skill. In reality this is a euphemism for a much broader set of excluded characteristics, some of which are more observable and concrete than others. One important individual characteristic in earnings determination is the specific type of university degree 5 Likelihood ratio tests confirm this. It should be noted that this procedure is legitimate contrary to the clain of Deb and Trivedi (1997), since the null hypothesis does not involve the boundary of the parameter space. 8 that the individual has obtained. Science, engineering and business degrees are now more likely to produce higher earnings profiles than those for historians or sociologists. Hence one of the reasons why earnings could have deteriorated is because these differentials are of recent origin or because the expansion of the educational system has favoured these social science and humanities programmes over more technical subjects. Of course, as many economists believe, the choice of degree programme is not unrelated to ability in its more abstract sense so that the deterioration in the quality of those pursuing a university degree which Côté and Allahar (2007) mention may also be related to the recent degree choices that have been made. When larger proportions of high school graduates attend universities it is very reasonable to expect that the upper tail of the ability distribution from which this population has been drawn has increased and that the average ability of university graduates has declined. While the plausibility of this conjecture is not in question further and more detailed research is needed to confirm this. The large number of components in the error terms suggests that there are many other potential contributors to individual earnings and to put most of the blame on the university system is perhaps premature. References [1] Bar-Or, Yuval, John Burbidge, Lonnie Magee, and A. Leslie Robb (1995), “The Wage Premium to a University Education in Canada.” Journal of Labor Economics, 13, 762-794. [2] Belzil, Christian (2006). “ The Return to Schooling in Structural Dynamic models”. IZA Working Paper #2370. [3] –––––– (2007) “Testing the Specification of the Mincer Equation”. IZA Working Paper #2650. To appear in The European Economic Review. [4] Belzil, Christian and Jörgen Hansen (2005). “A Structural Analysis of the Correlated Random Coefficient Wage Regression”. IZA Working Paper #1585. To appear in The Journal of Econometrics. [5] Boothby, Daniel and Torben Drewes (2006). “Post-Secondary Education in Canada: Returns to University, College, and Trades Education.” Canadian Public Policy, XXXII, 1-21. [6] Burbidge, John, Lonnie Magee, and A. Leslie Robb (2002). “The Education Premium in Canada and the United States”. Canadian Public Policy, 28, 203-217. [7] Cameron, Stephen and James J. Heckman (1998). “Life Cycle Schooling and Dynamic Selection Bias: Models and Evidence for Five Cohorts of American Males” Journal of Political Economy 106, 262-333. 9 [8] Card, David (1999). “The Causal Effects of Education on Earnings.” Chapter 30 in Handbook of Labor Economics, Volume 3, Edited by Orley Ashenfelter and David Card. Elsevier Science. [9] Chung, Lucy (2006). “Education and Earnings.” Perspectives on Labour and Income, Statistics Canada, Autumn, 28-36. [10] Côté James E., and Anton L. Allahar (2007). Ivory Tower Blues: A University System in Crisis. University of Toronto Press, Toronto. [11] Deb, Partha and Trivedi, Pravin K. (1997). Demand for Medical Case by the Elderly: a Finite Mixture Approach. Journal of Applied Econometrics 12: 313-336. [12] Green, David A. and W. Craig Riddell (2003). “Literacy and Earnings: An Investigation of the Interaction of cognitive and Unobserved Skills in Earnings Generation.” Labour Economics, 10, 165-184. [13] Heckman, James J., Lance J. Lochner, and Petra Todd (2006). “Earnings Functions, Rates of Return, and Treatment Effects.” Chapter 7 in Handbook of the Economics of Education, Volume I. Edited by Eric A. Hanushek and Finis Welch. Elsevier Science. [14] Heckman, James J. and Singer, B. (1984). A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data. Econometrica 52: 271-320 [15] Mare, R. D. 1980. Social Background and School Continuation Decisions. Journal of the American Statistical Association 75: 295-305 [16] Mas-Colell, Andreu, Michael D. Whinston, and Jerry R. Green (1995). Microeconomic Theory. Oxford University Press, Oxford, U.K. [17] McIntosh, James and Munk, Martin D. (2007). Scholastic Ability vs. Family Background in Educational Success: Evidence form Danish Sample Survey Data, Journal of Population Economics. 20 (1). [18] Morissette, René, Yuri Ostrovski, and Garnett Picot (2004). “Relative Wage Patterns Among the Highly Educated in a Knowledge-based Economy.” Paper 11F0019MIE No. 232 Business and Labour Market Analysis Division, Statistics Canada. [19] Sibley, Lisa, Sylvie Oullette, and Fernando Cartwright (2003). “Planning and Preparation: First Results from the Survey of Approaches to Educational Planning (SAEP) 2002”. Paper 81-595-MIE2003010. Culture, Tourism, and the Centre for Education Statistics Division, Statistics Canada. [20] Spence, A Michael (1973). “Job Market Signalling” Quarterly Journal of Economics, 87, 355-74 10 [21] Wedel, M., Desarbo, W.S., Bult, J.R. and Ramaswamy, V. (1993). “A Latent Class Poisson Regression Model for Heterogeneous Count Data” Journal of Applied Econometrics 8: 397-411. 11 Tables Table 1 Real Earnings Profiles And Employment Rates For University Graduates By Age. Real Employment Real Earnings Rate Earnings Age 1973 Employment Rate 1997 25 34479 0.623 39961 0.667 30 46225 0.941 43909 0.766 35 56208 0.930 62633 0.832 40 68539 1.000 72057 0.908 45 75487 0.929 68555 0.934 50 68367 0.913 61265 0.896 55 70398 0.944 84667 0.836 60 68765 0.857 57917 0.650 Average 60636 0.915 63952 0.821 Present Value 1239848 1128752 Sample Size 1133 2725 Table 2 Maximum Likelihood Parameter Estimates, 1973 Typ e Typ e 1 Typ e 2 Typ e 3 Unmixed α0 (a ) α1 (a ) α2 (a ) σ 10.247** (0.088) 10.452** (0.045) 10.163** (0.151) 10.326** (0.049) 0.090** (0.016) 0.056** (0.006) 0.080** (0.020) 0.063** (0.005) -0.013** (0.005) -0.012**(0.002) -0.017** (0.006) -0.013** (0.001) 0.121** (0.002) 0.256** (0.015) 0.664** (0.047) 0.400** (0.008) p 0.068** (0.024) 0.676** (0.041) 0.256** (0.039) 1.0 Variable Symb ol Intercept z z 2 Notes: For this table and table 3 †, ∗, and ∗∗ indicate significant at the 10, 5, and 1 p ercent levels, resp ectively. Standard errors are in round brackets. Table 3 Maximum Likelihood Parameter Estimates, 1997 Typ e Typ e 1 Typ e 2 Typ e 3 Unmixed 10.615** (0.034) 10.414** (0.052) 9.471** (0.378) 10.361** (0.050) 0.051** (0.016) 0.048** (0.006) 0.142** (0.037) 0.057** (0.005) Variable Symbol Intercept z z 2 α0 (a ) α1 (a ) α2 (a ) σ p -0.009** (0.001) -0.007**(0.002) -0.035** (0.010) -0.012** (0.001) 0.051** (0.007) 0.395** (0.023) 0.907** (0.085) 0.518** (0.007) 0.110** (0.014) 0.666** (0.050) 0.224** (0.052) 1.0 12 Table 4 Model Selection Criteria Number of Distributions Number of Parameters ln(L) ln(L) 1 2 -739.501 -2215.966 1 18 -568.025 -2060.693 2 23 -528.132 -1902.445 3 28 -522.805 -1855.439 0.293 0.163 Psuedo- R 2 1973 1997 Table 5 Real Earnings Profiles For University Graduates By Age, Type, and Year Typ e 1 Age Typ e 2 Typ e 3 Type 1 1973 Typ e 2 Typ e 3 1997 25 33478 39753 37281 44656 37266 23592 30 49645 49872 46334 54798 47041 36500 35 67411 57496 55473 66809 58816 49954 40 93035 67610 67416 75327 67903 59550 45 117814 73209 74796 74839 68672 58183 50 132210 70497 72604 78983 72938 53868 55 146372 67211 68207 78865 73422 43618 60 151303 61801 61306 72132 68419 30437 Share 0.068 0.676 0.256 0.110 0.666 0.223 Average Earnings 82921 59369 58158 70301 67150 49419 Present Value 1554000 1232000 1192000 1262394 1185738 894770 of Earnings 13