Student loans in Canada: an analysis of borrowing and repayment

Transcription

1 Economics of Education Review 21 (2002) Student loans in Canada: an analysis of borrowing and repayment S. Schwartz a,*, R. Finnie 1,b a Carleton University, School of Public Policy and Administration, Carleton University, Ottawa, Canada K1S 5B6 b Queen s University, School of Policy Studies, 99 University Avenue, Kingston, Ont., Canada K7L 3N6 Received 14 July 1998; accepted 1 September 2000 Abstract This paper reports the results of an econometric analysis of the borrowing and repayment patterns of Canadian bachelor s level university graduates, using data from the National Graduates Survey (NGS) of the class of After confirming the intuition that the level of borrowing is determined by supply-side rather than by demand-side factors, we analyze the repayment experience of the graduates. We calculate that the fraction of graduates who reported problems repaying their student loans was, overall, quite small, falling in the 7 8 percent range. Among both men and women, graduates with low current earnings and those in fields likely to have low lifetime earnings reported significantly greater problems with repayment. Holding other variables constant, women reported more difficulty in repayment than men. Overall, it would seem that women borrowed only slightly less than men, repaid as quickly as men (despite lower earnings), but reported having significantly more difficulty in repayment Elsevier Science Ltd. All rights reserved. JEL classification: I22 Keywords: Educational finance; Student financial aid 1. Introduction The extent to which Canadian university students pay for their own education changed substantially in the 1990s. At the end of the decade, students and their families were paying proportionally more, and governments proportionally less, than when the decade began. Almost all Canadian universities are publicly funded, relying on government grants and student tuition fees for the vast majority of their revenues. 2 In 1998, universities had a full-time equivalent enrolment of 644,434 students and general operating revenues, in current dollars, of * Corresponding author. Fax: address: saulfschwartz@carleton.ca (S. Schwartz). 1 Also at: Statistics Canada, Tunney s Pasture, Ottawa, Ont., Canada K1A 0T6. approximately $7.65 billion. 3 Government grants provided 64.5 percent of operating revenues while tuition fees accounted for another 30.6 percent. In 1979, the situation was different government grants comprised 83.8 percent of operating revenues and tuition fees only 13.3 percent. From 1979 to 1989, the 2 The other sources of revenues are investment income, bequests, donations and non-governmental grants. The phrase government grants, as used in the text, includes a relatively small government contracts component. 3 All dollar figures in this paper refer to Canadian dollars. In recent years, one Canadian dollar has been equivalent to between 65 and 70 US cents; thus multiplying each dollar figure by 2/3 will yield a number that roughly represents the value in US dollars. All years in this section are fiscal years; for example 1998 refers to the fiscal year spanning April 1, 1997 to March 31, /02/$ - see front matter 2002 Elsevier Science Ltd. All rights reserved. PII: S (01)

2 498 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) number of students (on a full-time equivalent basis) rose by 38 percent, creating the need for increased university operating expenditures. Real government grants did not keep pace, rising by just 11 percent. In an effort to cover the growing gap between operating expenditures and government grants, universities raised tuition fees, over this period, by 7 percent more than the Consumer Price Index. These trends continued into the 1990s. Overall university enrolments, on a full-time equivalent basis, grew by just under 4 percent in the period. At the same time, large federal and provincial budget deficits affected the level of real government grants. In fiscal 1989, those grants were $3.9 billion in 1986 dollars and hit a peak of $4.3 billion in fiscal By fiscal 1997, government grants had fallen to $3.6 billion, below their 1989 level. Since Canadian universities have no important revenue sources beyond government grants and tuition fees, the latter were bound to rise. 4 In the period, real tuition fees rose from $8.0 to $15.4 billion, an increase of 93 percent. The 1990s shift in the burden of university finances is clear. In real dollars, on a per student basis, government grants fell from $7569 in 1990 to $5543 in 1997, while tuition fees rose from $1363 to $2385 and expenditures fell by $800. Note, however, that full-time equivalent enrolments continued to grow despite the increased burden shouldered by students and their families and even though the number of 19- to 24-year-olds in 1997 was only 82 percent of what it had been in Between 1980 and 1997, enrolments grew by 42 percent even though real tuition fees grew by 289 percent. In the 1980s, when tuition fees were relatively low, the majority of university students were able to finance their education without direct government assistance. For potential students who needed financial aid because of low family income, grants were available from provincial programs. The major source of loans, for those who qualified by dint of low family income, was the federally funded Canada Student Loans Program (CSLP). Between 1984 and 1994, even though tuition fees were increasing, the maximum that could be borrowed from the federal CSLP remained constant. In 1994, the CSLP finally made more loan money available by increasing 4 One response to lower revenues was an attempt to reduce operating costs. In the 1980s, universities were able to reduce per student operating costs by 15 percent but the increase in the number of students led to an increase of 17 percent in overall operating expenses. Real university operating expenses per student then grew from $8108 in fiscal 1989 to $8646 in 1994 before falling back to $8288 in Little (1997) attributes the increase in operating expenditures to the rising seniority (and thus compensation) of university faculty and the difficulty, created by the tenure system, of laying off faculty to lower operating costs. the maximum that could be borrowed by more than 50 percent. 5 The provincial response to increased tuition fees varied, but the tendency was to replace provincial grant programs with provincial loan programs; that substitution allowed the same amount of subsidy dollars to put more money into the hands of students. One set of questions raised by these changes is normative. Who should pay for higher education? What is the share that should be shouldered by students themselves, by their parents and by the government? What are reasonable debt loads for graduates? Despite the importance of these normative issues, they are left to other discussions. In this paper, we address two important empirical questions concerning student borrowing and repayment by analyzing the experience of a sample of individuals who graduated with Bachelor s degrees from Canadian post-secondary institutions in The questions are: Did students borrow as much as they were offered during this period or, alternatively, did demand-side factors play an important role? 6 Which groups of borrowers, if any, were most burdened by their loan repayment? 2. Sample definition and the NGS databases The National Graduates Survey (NGS) and Follow-up databases include information on those who successfully completed programs of study at Canadian universities in 1982, 1986 and Information was gathered during interviews conducted two and five years after graduation. Response rates were around 80 percent for the first interview and 90 percent in the second, which is quite good for a survey of this type. Each of the three cohorts of the NGS has completed interviews for about 16,000 university graduates. The models estimated in this paper use 5 The maximum that could be borrowed was increased to $100 per week of full-time studies in It rose slightly to $105 per week in 1984 and then remained constant until 1994, when it was increased to $165 per week (Finnie & Schwartz, p. 11). A typical annual study period is 34 weeks. The maximum loan was therefore 34 times $165 or $5610. In , average tuition in Ontario was approximately $4000 for an academic year and living costs were about $7000. Thus the CSLP part of a maximum financial aid award covered tuition and about 20 percent of living costs. Provincial loans were also available. 6 Demand-side factors are those variables that are thought to play a role in the individual demand for post-secondary education, variables such as expected future earnings.

3 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) only the NGS cohort that represents those who graduated in The NGS data contain a wide variety of economic and demographic information about the graduates. 8 Most crucial to our work is that the data contain several key variables related to borrowing from federal and provincial student loan programs. 9 These include the amount owed at graduation, the amount still owed at the time of the first interview (conducted two years after graduation), and self-identified problems with loan repayment. The loan information, when combined with the economic and demographic information, makes the NGS data uniquely well suited for the study of student borrowing in Canada. 10 To create the files used in the analysis, we first limited the sample to those who received a Bachelor s degree in 1990 and who did not receive a second degree or study full-time in the subsequent two years. We then verified the key loan variables for consistency, and dropped or corrected a small number of records. 11 Finally, we deleted the relatively few observations with missing information. The variables used in the analysis are defined in Table 1, with corresponding sample means and standard deviations reported in Table A brief profile of bachelors graduates We begin by presenting a brief profile of the Bachelors graduates in the 1990 NGS cohort. We report their 7 Analyses of the other NGS cohorts can be found in Finnie and Garneau (1996a,b) and in Finnie and Schwartz (1996). The NGS also collected information on graduates from post-secondary institutions other than universities. 8 The NGS sampling frame was stratified by province, level of education and field of study. In all of the models estimated below, we use the associated sampling weights. 9 During the 1980s, the bulk of student borrowing consisted of loans guaranteed by the CSLP. Provincial loan programs were quite small at that time, although they grew considerably in the 1990s. 10 Several limitations of the NGS data should be noted. First, only data from the first interview were available for the 1990 graduates so that our analysis was limited to a relatively short post-graduation period. Nonetheless, most student loan default (as well as a significant amount of loan repayment) occurs in these early years. Second, the experience of those who borrowed but who never graduated from a program is not captured by the NGS and that group may be quite different in their borrowing and repayment experience. Finally, the NGS did not collect information about borrowing for purposes other than schooling such as credit card balances, car loans and home mortgages. 11 We examined about 80 observations, corrected approximately one quarter of those, dropped another quarter and left the rest unchanged. distribution by gender and field of study and their postgraduation earnings and unemployment rates. Table 3 shows the percentages of men and women in each field of study. Engineering was dominated by men (17 percent of the male graduates versus just 2 percent of the females) as was Mathematics and Physical Sciences. The Medical and Health professions (excluding doctors) were dominated by women. Commerce, Economics and Law were disproportionately male while Education, Fine Arts and Humanities, and Other Social Sciences were disproportionately female. Only doctors had similar proportions of male and female graduates. 12 Post-graduation patterns of earnings and unemployment rates are important for our analysis. Table 3 shows mean earnings, by gender and field of study, for those who were working full-time two years after graduation. The first result is unsurprising men earned more, overall, than women. This is true partly because men earned more than women within a given field of study (with the only exceptions being Engineering and Fine Arts and Humanities) and partly because women were more concentrated in fields with relatively low average earnings (such as Other Social Sciences and Fine Arts and Humanities). The second result is that there were clear earnings patterns by field. The ranking of average earnings by field of study was broadly similar for men and women. Doctors had considerably higher average earnings than any other group. Those in Engineering, Other Medical and Health, Commerce, Economics and Law, and Mathematics and Physical Sciences had above average earnings. Those in Education, Fine Arts and Humanities, Other Social Sciences, and Agriculture and Biological Sciences had below average earnings. The last two columns of Table 3 show unemployment rates two years after graduation, by gender and field. Overall, men had a slightly lower unemployment rate than women. Within fields of study, however, there was no general advantage for men; thus, the overall difference was principally due to differences in fields of study among male and female graduates. We also see that, among men, the fields with the highest average earnings tended to be those with the lowest unemployment rates; for women, the pattern was roughly similar, although not quite as well defined. Two of the patterns in Table 3 will be especially relevant to our later findings 12 Throughout the paper, we use capital letters when we are referring to the name of one of the field of study categories (e.g. Commerce, Economics and Law) and use lower-case letters to refer to graduates from a particular field of study (e.g. lawyers or doctors).

4 500 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 1 Definition of variables Borrowed from loan program Respondents who had an outstanding student loan balance upon graduation were coded as 1 ; otherwise, individuals are coded as 0 Amount owed at graduation For those who borrowed, the dollar value of student loans outstanding at graduation, in constant $1990. Defined only for those with outstanding loan balances at graduation Proportion repaid The proportion of the amount owed at graduation that had been repaid as of the first NGS interview, two years after graduation. Calculated as one minus the ratio of the amount owed at the first interview to the amount owed at graduation. Defined only for those with outstanding loan balances at graduation Difficulty with repayment Respondents who reported having difficulties with the repayment of their student loans were coded as 1 for this variable; otherwise, the variable was coded as 0. Defined only for those with outstanding balances as the time of the first interview Gender Male=1; Female=0 Annual earnings in 1992 Annual earnings, in $1990, at the time of the 1992 NGS interview, two years after graduation Field of study Standard USIS categories, with some re-groupingbased on preliminary analysis of the data. In particular, Economics graduates were taken from Social Sciences and put with Commerce and Law graduates; Medicine graduates were separated from Other Medical and Health. There is one variable per field of study coded 1 for respondents in that field and 0 otherwise Age Respondent s age at the 1992 NGS interview Province/region In the model of whether or not the respondent borrowed from a loan program, this variable refers to the province or region of residence before enroling in the respondent s post-secondary program. In the repayment models, the variable refers to residence at the time of the interview. Preliminary analysis resulted in treating the Atlantic provinces as a single region. There is one variable per province/region, coded 1 for respondents in that province and 0 otherwise Parental education Respondents reported the highest level of education attained by both their mother and their father, as applicable. Four education levels were defined: less than high school graduation; less than a Bachelor s degree (BA), including high school graduation; and BA or higher. For both fathers and mothers, a separate 0 1 indicator was then created for each education level. The variable was coded 1 if the respondent s parent s education was in that category and 0 otherwise Marital status Respondents reported their marital status at the time of the 1992 NGS interview. Three 0 1 indicators were then created: (1) never married; (2) currently married or living in a commonlaw arrangement (and not separated from the partner); (3) formerly married (including those who were separated, widowed or divorced). The indicators took the value 1 if the respondent reported a marital status in the relevant category and was coded 0 otherwise. Children present Those with children present in the household were coded as 1 ; others were coded as 0 Schooling characteristics Respondents reported on various aspects of their post-secondary schooling experience. Five 0 1 indicators captured the following experiences: (1) moving from one province to another in order to enrol in a post-secondary program; (2) attending a post-secondary program on a parttime basis at some point; (3) attending a post-secondary program that involved a co-op placement; (4) working at least three years prior to enroling in their post-secondary program. Each variable was coded as 1 if the respondent reported the characteristic and was coded 0 otherwise. In addition, a variable indicating the normal length of the post-secondary program, measured in years, was also included. male graduates had generally higher earnings and, to a lesser degree, lower unemployment rates than female graduates; with some relatively minor variation, there was a clear ranking of fields of study in terms of their average earnings and unemployment rates. 4. Do demand-side factors influence student loan borrowing? In this section, we test our first hypothesis that student borrowing was constrained by the supply of loans rather than by the demand for them. We begin with a very brief summary of the well-known human capital model of the individual demand for education and then describe the institutional context of student borrowing in Canada. We then turn to two empirical results relating to the question of whether factors influencing the demand for loans affected student borrowing The human capital model The rationale for student borrowing is well known. At the level of individual decision-making, the human capi-

5 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 2 Means and standard deviations of variables used in the econometric models Male Female Mean SD Mean SD Borrowed from loan program Amount owed at graduation $8800 $5880 $8760 $5650 Proportion repaid Difficulty with repayment Annual earnings ($1990) $33,620 $19,190 $30,710 $14,140 Age Province/region Atlantic Quebec Ontario Manitoba Saskatchewan Alberta British Columbia Father s education Less than high school Less than completed BA BA or higher Not stated Mother s education Less than high school Less than completed BA BA or Higher Not stated Marital status Never married Currently married Formerly married Children present Schooling characteristics Migrated to go to school Part-time at some point In co-op program Re-entered after working Normal length of program tal model views post-secondary education as an investment that yields financial (and non-financial) returns over the lifetime of the individual. In every time period, each individual decides whether to work full-time or to go to school by comparing the net present of value of going to school to the net present value of working. The cost of post-secondary education consists largely of foregone earnings, although that cost may be relatively low in times of high youth unemployment. Moreover, as we saw in the last section, a direct cost tuition fees has risen substantially in recent years. To finance those costs, the rate of interest on loans must be considered (whether the rate applies to borrowed funds or to funds that might otherwise have been invested) The institutional context Institutionally, the student loan process in Canada involves three steps. First, a student planning a program of full-time studies for an upcoming period must decide whether or not to apply for a student loan. For those who apply, the next step is taken by provincial authorities. Combining information provided by the applicant with standardized need assessment procedures, the province:

6 502 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 3 Percentage of male and female graduates, mean 1992 earnings and unemployment rates by field of study Gender (%) Average earnings ($1990) Average unemployment rate (%) Male Female Male Female Male Female All fields of study $33,620 $30, Education $32,560 $30, Fine Arts and Humanities $23,660 $26, Commerce, Economics and Law $35,540 $31, Other Social Science $31,550 $27, Agricultural and Biological Sciences 4 6 $29,380 $27, Engineering 17 2 $34,540 $37, Doctors 4 4 $48,530 $47, Other Medical and Health 0 9 $37,600 $34, Mathematics and Physical Sciences 10 3 $34,220 $31, All dollar values are expressed in Canadian dollars; see Footnote 2 (a) determines whether the student is eligible for a loan; and (b) calculates the dollar amount of loan to be offered to those deemed eligible. The third step is taken by the student who decides whether to accept the loan (if one is offered) and whether to borrow all or only part of the amount offered. Because the goal of the government subsidy (and government loan guarantees provided to private-sector lenders) is to aid students with demonstrated financial need, ability to repay is not a consideration in determining the size of the loan offered to each applicant. After students have left full-time studies, several programs exist to help those having difficulty repaying their loans. For example, the federal CSLP has an interest relief program that pays the interest on the loans for up to 54 months while borrowers make no payments. For those graduating in 1990, however, only 30 months of interest relief were available. Because of these special features features that would make a commercial loan extremely attractive to borrowers one hypothesis is that all eligible borrowers borrow as much as they possibly can. If this hypothesis were true, the observed data on borrowing would reflect only the rules that govern the student loan system (or, more correctly, the rules as they operate in practice). An alternative hypothesis is that demand-side factors, such as expected future earnings, play an important role. In our context, looking only at those who have graduated from university, the implication is that some students borrow less than they could because they fear that they will not be able to repay the loans. If so, students in remunerative fields of study (such as engineering or business) would be more likely to borrow than students in less vocationally oriented fields of study (such as fine arts). Moreover, since women have lower expected future earnings than men, they might be less likely to borrow. In theoretical terms, this hypothesis is that some students foresee that the expected rate of return to education may not exceed the certain cost of borrowing Empirical results of borrowing model The amount that students borrow is constrained to be less than or equal to the amount offered by provincial authorities which is determined by exogenous factors such as family income and the parameters of the student loan program. Moreover, in a survey of CSLP borrowers conducted in the spring of 1997, 92 percent of respondents reported accepting the full amount of the loan offered to them. This leads us to assume that, once the decision to apply for a loan has been made, the amount borrowed is exogenously determined. Demand and supply factors might still affect the decision to apply for a loan and that decision is the focus of the analysis presented next. Using the NGS data, we use a simple probit model to analyze whether or not a graduate had ever borrowed, over the course of his or her post-secondary education, from a federal or provincial student loan program. The explanatory variables represent both demand and supply factors. 13 Field of study and gender play a central role in our analysis. They are the primary candidates for demand- 13 As a check on our assumption that the amount borrowed was determined exogenously, we estimated a two-equation model of having a loan and the amount borrowed. This nowstandard model, pioneered by James Heckman, adjusts for the selection bias that might be created by unobserved variables being correlated with included independent variables in the amount borrowed equation. We found no evidence that demand-side factors play a role in determining the amount borrowed. Moreover, the sample selection correction term was not significantly different from zero. Results are available from the authors.

7 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 4 Proportion of graduates with loans and average amount owed (for borrowers) at graduation by field of study Proportion with loans (%) Average amount owed at graduation ($1990) Male Female Male Female All fields of study $8800 $8760 Education $9690 $9160 Fine Arts and Humanities $8570 $7120 Commerce, Economics and Law $9210 $8970 Other Social Science $7610 $7790 Agricultural and Biological Sciences $8930 $9010 Engineering $7830 $8780 Doctors $12,820 $12,120 Other Medical and Health N/A $9170 Mathematics and Physical Sciences $7750 $5640 All dollar values are expressed in Canadian dollars; see Footnote 2 side influences on the decision to borrow from a student loan program because they capture the effects of differences in expected future earnings. Gender differences in borrowing might also reflect unobserved male female differences in attitudes toward debt still a demand factor, but separate from the effect of expected incomes. The primary supply-side variable is province of residence. Other variables included in the borrowing model generally affect both supply and demand and enter as controls. They also provide a fuller description of the factors which affect borrowing and include parental education, age, and the other characteristics defined in Table 1. Two columns of Table 4 show the proportion of Bachelors graduates in our sample who finished their studies with a student loan. Looking at Table 4, we see that just under half of the graduates had borrowed from student loan programs. Overall, men were more likely to have borrowed than women but there is no clear gender pattern within fields of study. Nor is there a consistent relationship, in these cross-tabulations, between the patterns of borrowing in the various fields and the corresponding average earnings and unemployment rates. For example, Table 4 indicates that graduates in Engineering, a high earnings/low unemployment field, had relatively high rates of borrowing. Graduates from Commerce, Economics and Law, however, had relatively high earnings but relatively low rates of borrowing. Cross-tabulations do not tell the whole story, however, because graduates vary along dimensions other than gender and field of study, and these differences might affect the probability of borrowing. The first two columns of Table 5, therefore, show the coefficient estimates from a multivariate probit model of the probability of graduating with a student loan, estimated separately for men and women. We focus first on the implications of the findings for our analytic hypothesis that those in fields with high average earnings and low employment rates will borrow more than those in fields with low average earnings and high unemployment rates. To illustrate the findings, the last two columns of Table 5 present the results of simulations, based on the parameter estimates, that show how the probability of borrowing is estimated to change with changes in the independent variables. We begin by calculating the probability of borrowing for prototypical (or baseline ) male and female graduates. 14 Because the characteristics of the baseline male and female graduates are the same, any differences in the predicted probabilities are due to differences in the coefficients. The baseline probabilities of borrowing 0.27 for men and 0.24 for women are shown in the first row of the last two columns of Table Next, we successively alter one characteristic at a time to generate a full set of predicted probabilities based on the corresponding parameter estimates. 16 For example, the coefficient estimate for each field of study is used to predict a field-specific probability of borrowing, holding 14 The prototypical borrowers were 28 years old, lived in Ontario and had parents who both had bachelors degrees or higher. Their major field of study was other social science. They had not migrated out of Ontario to go to school, were full-time students, were not in a co-op program, had not reentered school after working. Their normal period of study was four years. 15 The baseline probabilities are considerably lower than the probabilities calculated at the means of the independent variables because the baseline individuals are assumed to have two parents with relatively high levels of education (Bachelor s or higher), which tends to make individuals less likely to be eligible for loans. 16 Results are not shown for variables other than gender, field of study and province.

8 504 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 5 Probit model of student loan borrowing Coefficient estimates (std. errors) Predicted probabilities Male Female Male Female Predicted probability of borrowing Age (0.003) (0.048) % % Field of study Education (0.034) (0.025) 0.38* 41% 0.31* 30% Fine Arts and Humanities (0.036) (0.028) % 0.32* 35% Commerce, Economics and Law (0.029) (0.027) % 0.26* 8% Agricultural and Biological Sciences (0.047) (0.037) 0.30* 12% 0.34* 43% Engineering (0.033) (0.058) 0.33* 21% 0.31* 29% Doctors (0.052) (0.046) 0.38* 39% 0.46* 90% Other Medical and Health (0.100) (0.032) % % Mathematics and Physical Sciences (0.037) (0.002) 0.24* 11% 0.25* 4% Province/region Atlantic (0.035) (0.029) 0.56* 107% % Quebec (0.023) (0.021) 0.42* 55% % Manitoba (0.043) (0.039) 0.32* 18% 0.34* 40% Saskatchewan (0.041) (0.037) 0.36* 32% 0.34* 42% Alberta (0.032) (0.030) 0.50* 85% 0.48* 99% British Columbia (0.033) (0.033) 0.33* 22% 0.29* 20% *Prediction based on a statistically significant coefficient. all other characteristics of the baseline individual fixed, thus isolating the effect of the field. We then compare each predicted probability with the baseline probability and express the difference in percentage terms. These percentage differences appear beside the predicted probabilities. Predictions based on parameter estimates that were significantly different from zero at the 1 percent level of significance are marked with an asterisk. For example, the baseline probability for women assumes that the woman s major field of study was Other Social Sciences; if that hypothetical baseline woman were instead in Education, her predicted probability of borrowing would rise from 0.24 to 0.31 and that prediction is based on a statistically significant coefficient estimate (see column 5 of Table 5). The predicted probability of 0.31 is 30 percent higher than the baseline probability of 0.24 (see column 6 of Table 5). This simulation technique, which we will use again in later sections, provides two useful perspectives on the results: we can compare the probability of borrowing for men and women for any given set of characteristics, and we can observe the effect of each variable on the probability of borrowing for each group separately. As an example of the first perspective, Table 5 indicates that, holding other factors constant, female doctors have a predicted probability of borrowing of 0.46 and were thus more likely to borrow than otherwise similar male doctors, who have a predicted probability of From the second perspective, Table 5 predicts that female doctors were more likely to borrow than female engineers, because their estimated probability of 0.46 is higher than the female engineers predicted probability of Does field of study affect the probability of borrowing? Looking across the results reported in Table 5 both the probit coefficients and the predicted probabilities derived from them we again see no clear patterns by field of study. 17 There is no compelling evidence that borrowing rates were positively related to expected future incomes. Both for men and for women, there are generally lower predicted borrowing rates among graduates with degrees in Other Social Sciences (low expected earnings according to Table 3), Commerce, Law and Economics (high expected earnings), Other Medical and Health (high), and Mathematics and Physical Sciences (high). There are generally higher predicted probabilities of borrowing for those in Fine Arts and Humanities, and Agricultural and Biological Sciences (low expected earnings) and Engineers (high expected earnings). The highest predicted rates of borrowing are for doctors (who 17 An exception is that doctors, whether male or female, were more likely to have borrowed than other groups.

9 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) have the highest earnings in Table 3) and for graduates in Education (who have earnings near the overall average) Does gender affect the probability of borrowing? Table 4 reveals an overall difference of 4 percentage points between the proportion of male and female graduates who had student loans at graduation; 48 percent of male graduates had loans as compared to 44 percent of female graduates. Looking across fields of study in Table 5, however, we see no clear tendency for men to borrow more than women. Women have higher predicted probabilities of borrowing in four of the nine fields reported in Table 5. We conclude that the greater expected earnings of men do not generally lead them to be more likely to borrow The effect of other variables on the probability of borrowing Province is taken to be an important proxy for supply-side factors such as differences in how provincial student aid programs interacted with the CSLP and provincial differences in educational costs. The reason for these provincial differences lay largely in different packaging of loans and grants across provinces. For example, Ontario had a substantial grant program in the late 1980s and loans were offered only to students whose need had not been met by the grant program. In some other provinces, loans were the first type of aid offered, so that borrowing was more prevalent. Table 5 shows that province was an extremely important determinant of the probability of borrowing. Graduates from Atlantic Canada were the most likely to have borrowed, followed closely by Alberta and then Quebec. Graduates from Manitoba, Saskatchewan and British Columbia had the next highest probabilities of borrowing while Ontario graduates were the least likely to have borrowed. Furthermore, some of the effects are quantitatively large; for example, baseline individuals from the Atlantic region had a predicted probability of having a student loan that was more than double (109 percent higher) that of an otherwise similar individual from Ontario. In summary, our results do not suggest that demandside factors such as expected future earnings, represented here by field of study and by gender, played any systematic role in determining the probability of graduating with a student loan. We therefore conclude that student loan borrowing was almost entirely constrained by supply-side factors in the late 1980s and that most individuals borrowed when they were eligible to do so. 5. Loan repayment and difficulties encountered in repayment In this section, we turn to our second broad question, which concerns the burden of student loan repayment. We first summarize the past literature on this topic (with a more extensive review in Appendix A). We then present two analyses of the burden of debt repayment using the NGS. Paying back a loan clearly imposes a financial burden on the borrower. The relevant question is whether that burden is so great as to threaten the financial well being of the borrower. Because student loans are offered to young people without any consideration of their future ability to repay, fears about excessive borrowing have been voiced since North American student loan programs were introduced in the 1950s and 1960s. Efforts to measure how many student loan borrowers face financial hardship as they attempt to repay their loans have had to deal with two difficult issues. First, any definition of financial hardship is inherently subjective. As a result, in the words of Hansen (1991, pp ) Researchers have never been able to agree on what level of debt is likely to be manageable for those who borrow to pay for college; given this lack of consensus, it becomes difficult for policymakers to know if and when the point of too much borrowing is reached. Despite the increased importance of student loans in the financing of higher education, both in the US and Canada, there remains no consensus on the proportion of income that can be allocated to student loan repayment. 18 Second, student loan debts are only one of a number of debts by which young people might be burdened but very few data sets are rich enough to contain all debts that a former student might have incurred. For example, many young people use credit cards and might therefore have outstanding balances on them. Individuals with identical ratios of student loan payments to income might therefore have quite different debt repayment burdens. Without data on all debt obligations, no measure of financial hardship can be complete. A summary of the previous literature is that despite long-standing and continuing concern about excessive debt burdens, students have continued to borrow. The proportion classified as facing a severe burden by researchers is generally less than 25 percent of all borrowers. Thus, any dire predictions of the consequences of increased borrowing should at least be leavened with 18 According to Baum and Saunders (1998, p. 4): there has been no official benchmark set recently by the financial aid community in 1986 the National Association of Student Financial Aid Administrators (NASFAA) said that anything beyond an 8% monthly student loan payment would begin to be burdensome. Today it is frequently suggested that 10 12% is likely to cause hardship.

10 506 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) the knowledge that past fears of widespread and dramatic problems have not been realized Analyzing the NGS data on the proportion of debt repaid within two years of graduation Under the rules in force in 1990, borrowers who were not intending to continue in a full-time program of study entered a six-month grace period as soon as their studies were finished. During that period, while the interest payments on their loans continued to be paid by the government, students were supposed to consolidate their loans at a single financial institution and arrange a payment schedule. The typical repayment schedule for CSLP loans was a fixed period of ten years so that the rate of repayment was, to some extent, institutionally determined. 19 The standard theoretical model for analyzing the repayment pattern that is desired by borrowers (as opposed to that required by lenders) asserts that borrowers will want a repayment pattern that equates the ratio of the marginal utility of consumption in any two time periods to the relative price of money in the two time periods. 20 For example, in a two-period model, the subjective value of a dollar consumed today is compared to the monetary value (including interest) of saving it until tomorrow. In intuitive terms, this means that the repayment patterns desired by borrowers will depend on how much they value current consumption. That value, in turn, will vary widely among young people just leaving university. For example, some will have high current needs or desires; this would include those with a low current income and those who are saving for big ticket items such as houses or cars. Such individuals will prefer to repay their loans relatively slowly. For others, their current income will exceed their current needs and wants this includes those who find outstanding debt distasteful and who are therefore willing to forego current consumption to repay their loans. Such individuals will prefer to repay their loans relatively quickly. Of particular interest to us is the case of former stu- 19 A lender whose collateral is the asset purchased with the loan will want to set a repayment pattern that assures that the value of the collateral exceeds the outstanding loan balance. For assets that depreciate quickly (such as motor vehicles), lenders will want rapid repayment; for assets that do not depreciate quickly (such as residential properties) repayment can be slower. Human capital cannot be repossessed so these rules cannot be easily applied. 20 See, for example, Cameron (1984, Chapter 6). The idea is that the marginal rate of time preference (the slope of an intertemporal indifference curve in a two period model) must be equal to the relative price of money (the slope of the intertemporal budget constraint). dents who expect high earnings in the future. On the one hand, such individuals might want to smooth their consumption over time by repaying their loans more slowly in the early, lower-than-average years immediately following university. On the other hand, those with high future earnings might also have high current earnings and thus be able to repay more quickly if they so desire. Empirically, it is clear that not all former students repaid their loans at the same rate. In our sample of graduates, as shown in Table 6, we estimate that about 9 percent made no payments. 21 At the other end of the repayment spectrum, borrowers could choose to repay more quickly than required. The overall number of CSLP borrowers who did so is unknown, but Table 6 indicates that over 20 percent of the NGS Bachelors graduates reported repaying their loans in full within two years of graduation. There was, therefore, considerable variation in the proportion of the outstanding loan balance repaid in the two years between graduation and the first NGS interview. Despite the theoretical uncertainty as to the desired repayment stream, we expect repayment rates to be a function, in general, of the graduates ability to repay. The most important indicators of that ability are the size of the loan, and both current and expected post-graduate earnings. Because the proportion repaid was constrained to lie in the 0 1 interval, we employ a two-sided Tobit model to estimate its determinants. The independent variables differ somewhat from those included in the model of the probability of borrowing. We add the amount owed at graduation and current earnings, as well as marital status (as of the first interview) and a 0 1 indicator of the presence of children. Province of residence at the first interview replaces the province of origin. Estimated coefficients for the model are shown in the first two columns Table 6 Distribution of proportion of loan repaid Male Female All graduates % 100% 100% 21 We infer that graduates made no payments if the amount outstanding at graduation was the same as the amount outstanding at the first interview, two years after graduation.

11 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 7 Tobit model of proportion repaid Coefficient estimates (std. errors) Predicted probabilities Male Female Male Female Predicted proportion repaid Age (0.002) (0.001) % % Amount of loan (0.001) (0.001) 0.47* 4% 0.41* 5% Current earnings (0.000) (0.000) 0.50* 3% 0.45* 3% Field of study Education (0.019) (0.015) % % Fine Arts and Humanitites (0.022) (0.017) % 0.49* 12% Commerce, Economics and Law (0.017) (0.016) % 0.50* 15% Agricultural and Biological Sciences (0.028) (0.022) 0.56* 14% 0.54* 23% Engineering (0.019) (0.032) 0.56* 13% 0.66* 53% Doctors (0.028) (0.025) 0.60* 23% 0.57* 30% Other Medical and Health (0.059) (0.020) % 0.58* 35% Mathematics and Physical Sciences (0.022) (0.029) 0.52* 6% 0.56* 29% Province/region Atlantic (0.018) (0.017) 0.44* 10% 0.34* 21% Quebec (0.013) (0.012) 0.45* 9% 0.28* 35% Manitoba (0.028) (0.026) % % Saskatchewan (0.026) (0.025) % % Alberta (0.016) (0.015) 0.53* 8% % British Columbia (0.019) (0.018) 0.54* 10% 0.40* 8% of Table To illustrate our results, we again employ the technique utilized in Section 4. We use the Tobit coefficients to calculate predicted probabilities for male and female baseline individuals and then, in the last two columns of Table 7, we show the predicted probabilities associated with a given change in each independent variable. The changes simulated in the last four columns of Table 7 are: a one year increase in age, a $5000 increase in earnings, and a $1000 increase in the amount of the student loan. As we would expect, the proportion repaid was positively related to current earnings. For both men and women, an additional $5000 in earnings increased the predicted proportion repaid by 3 percent. The proportion repaid was negatively related to the amount owed at graduation. For both men and women, a $1000 increase in the total amount owed at graduation reduced the proportion repaid by 4 5 percent. Also of interest here is the similarity in the overall repayment rates of men and women. The average proportion repaid was 0.45 for men and 0.44 for women (Table 2). Moreover, Table 6 shows that the distribution of the proportion repaid was quite similar for men and women. Table 2, however, indicates that the average cur- 22 Again, only a subset of the coefficients is presented in the Table. rent earnings of men were higher than that of women by about $3000 per year. In Table 7, the estimated coefficient on current earnings is for both men and women. Holding other factors constant, the model thus predicts that men would repay a higher proportion of their loans. How then can we account for the similarity in repayment rates when average earnings are different? The answer to this question might lie either in different average values of the other variables or in differences in the coefficients between men and women. In the context of our econometric model, we address the question by following the now-standard procedure of decomposing the overall gender difference into a part due to the male female differences in the effects of the variables included in the models (that is, to differences in the coefficient estimates reported in Table 7) and the part due to male female differences in characteristics (that is, to differences in the mean values of the regressors). With the current models, however, this decomposition does not provide any conclusive answers. For example, there are significant differences between men and women in the proportions who graduated from various fields of study (Table 2) and differences in the estimated coefficients on the field of study indicators (Table 7). Both of these differences affect repayment rates. But men and women also differ in other characteristics and in other estimated coefficients, and no general pattern emerges.

12 508 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 8 Monthly loan payments as a percent of monthly post-tax earnings, by loan level and earnings categories. Monthly payment levels Earnings Pre-tax fulltime Post-tax fulltime $75 per month $150 per month $224 per month earnings earnings Payment as a percent of post-tax monthly earnings 25th Percentile $26,005 $20, % 8.60% 12.90% 50th Percentile $33,350 $25, % 7.00% 10.49% 75th Percentile $39,604 $29, % 6.09% 9.14% To compute the earnings quartile, we used the reported wages of those NGS respondents who reported working full-time at the time of the survey. To compute net monthly earnings, the reported wage was first converted into annual earnings. Then, annual earnings, net of taxes, were computed by using published tax schedules as they would apply to a hypothetical unmarried adult without children. The monthly loan payment amount was computed using loan amounts of $5000, $10,000 and $15,000, amounts that correspond roughly to quartiles of amount borrowed. All dollar values are expressed in Canadian dollars; see Footnote 2. Our model thus shows that the overall repayment rate for women is similar to that for men despite their lower current earnings, with that overall equality being the result of a variety of effects related to men s and women s characteristics and repayment behaviour Analyzing the NGS data on the difficulty of student loan repayment This section attempts to measure the determinants of the debt burden created by student loans. We should begin, however, by putting the incidence of repayment problems in context. More than half of the 1990 Bachelor s graduates finished their studies free of government loans and thus faced no repayment burden at all. Furthermore, just under 25 percent of those who borrowed roughly 12 percent of all graduates had fully repaid their loans within two years of graduation and thus may be presumed not to have been overly burdened by their debt. In the end, only about 35 percent of all graduates in our sample had outstanding student loans at the time of the first NGS interview in Monthly payments as percentage of full-time earnings For those who were working full-time, it is instructive to compute the percentage of post-tax monthly earnings that must be devoted to student loan payments. Table 8 is divided into nine cells. Each cell represents a combination of a level of post-tax earnings and a level of debt repayment. The levels of post-tax earnings are the quartiles of full-time earnings reported by NGS Bachelors graduates. The levels of monthly loan payments are derived from three levels of cumulative debt $5000, $10,000 and $15,000 that are rough approximation to the quartiles of total outstanding debt in the NGS. Table 8 shows that graduates who had earnings at the 25th percentile and outstanding loans at roughly the 75th percentile would use about 13 percent of their net monthly income to service their student loan. This is near the percent threshold mentioned above. This low earnings/high payment group is the only one in the table, however, that would have repayment burdens near that threshold. The other groups have burdens that are considerably lower. Before one concludes that very few graduates are heavily burdened by their student loan repayment, we note that the repayment burden was undoubtedly heavier for graduates who were either working part-time or not at all. Moreover, the problem that other debt payments are not recorded on the NGS remains potentially important Self-reported debt burden Another way to estimate the extent to which graduates were experiencing heavy repayment burden is to analyze the extent of self-reported problems. The NGS asked the group who had outstanding student loans if they had experienced difficulty with the repayment of their loans. 23 Overall, 19 percent of the male graduates and 26 percent of the female graduates responded that they had indeed been having trouble (Table 9). Applying those two proportions to the 35 percent who had outstanding student loans, we infer that the proportion of all Bachelor s graduates who experienced difficulties 23 Self-reported problems with repayment refers to responses to a question asked of survey respondents. We do not know whether those reporting problems actually were behind in their payments or if they had defaulted on their student loans. Student loan default, in 1990, occurred when the lending institution reported a series of missed payments, requested reimbursement of the loan principal and interest and received that reimbursement.

13 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 9 Proportion reporting repayment difficulties (for those still owing) Male Female All fields of study Education Fine Arts and Humanities Commerce, Economics and Law Other Social Science Agricultural and Biological Sciences Engineering Doctors Other Medical and Health 0.08 Mathematics and Physical Sciences with the repayment of student loans was in the range of 7 8 percent. 24 The next question is whether particular subgroups of graduates are especially likely to be experiencing difficulty in repaying their loans. Using the presence of selfreported difficulties as the measure of difficulty in repayment, this section begins by looking at the prevalence of self-reported problems by major field of study. Then we present the results of a multivariate probit model of the probability of reporting difficulties with student loan repayment. The frequency of repayment problems followed a clear pattern by major field of study. In particular, fields characterized by low average current earnings had higher proportions of graduates reporting difficulty (Table 9). For example, the three fields with the lowest average earnings (Fine Arts and Humanities, Other Social Sciences, and Agricultural and Biological Sciences) were also those with the highest incidence of repayment problems. This pattern remained when we estimated a multivariate probit model of the probability of reporting difficulty in repayment. The estimated coefficients for this model 24 That statement requires some caveats, however. First, our research looked only at graduates, and we expect that those who incurred loans but did not complete their programs may have been prime candidates for repayment problems. Furthermore, some of those who repaid their loans rapidly may have done so only with difficulty. Finally, as previously discussed, the most recent graduates covered by our data are those of 1990; it is virtually certain that the financial situation of graduates has become generally more difficult since that time and, therefore, that repayment problems have become more widespread. are shown in the first two columns of Table 10 while the last two columns show the predicted probabilities of reporting problems. We find that graduates in fields with lower expected future earnings have a higher predicted probability of experiencing repayment problems, even after controlling for current earnings and the amount owed at graduation. Furthermore, the differences by field are not only statistically significant in most cases, but also quantitatively important, with some fields having predicted probabilities of repayment problems almost double the size of others. As expected, differences in the amount owed at graduation were statistically significant and had a large impact on the probability of reporting difficulties. An extra $1000 in loans increased the probability of repayment problems by 5 percent for women and 7 percent for men. Differences in current earnings also had effects that were statistically significant and in the expected direction: higher earnings meant lower predicted probabilities of repayment problems. The sizes of those effects, however, were quite different for men and women. The coefficient estimate for women was ten times the size of the coefficient for men. As a result, a $5000 increase in earnings is predicted to lead to a 2 percent decline in the incidence of repayment problems for men but a 17 percent decline for women. Overall, the predicted probabilities indicate that female graduates reported more problems than male graduates for a given set of characteristics, including field of study. Moreover, if we decompose the overall gender difference into differences due to coefficients and differences due to characteristics, we find that differences in the coefficients account for more than the entire male female difference in repayment problems A summary of findings, some implications, and directions for further research This paper reports the results of an econometric analysis of the borrowing and repayment patterns of recent Bachelor s level university graduates using data from the NGS of the class of The study exploits the unique combination of information concerning student loans, educational experiences, and early labour market outcomes available in the NGS data. We ask two questions of the data. First, did demandside factors such as gender or field of study play an important role in determining the probability that 25 The overall advantage of 6.4 percentage points for men decomposes into an advantage for men of 7.9 percentage points due to coefficients versus a smaller and offsetting advantage of 1.5 percentage points for women due to their average characteristics.

14 510 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Table 10 Probit model of difficulties with repayment Coefficient estimates (std. errors) Predicted probabilities Male Female Male Female Predicted proportion repaid Age (0.007) (0.005) 0.14* 7% 0.32* 6% Amount of loan (0.003) (0.003) 0.14* 7% 0.32* 5% Current earnings (0.001) (0.001) 0.13* 2% 0.25* 17% Field of study Education (0.070) (0.052) 0.09* 29% % Fine Arts and Humanities (0.077) (0.058) 0.25* 91% 0.40* 30% Commerce, Economics and Law (0.063) (0.057) % % Agricultural and Biological Sciences (0.103) (0.075) % % Engineering (0.072) (0.121) % 0.42* 37% Doctors (0.119) (0.116) 0.05* 63% 0.13* 59% Other Medical and Health (0.088) 0.14* 54% Mathematics and Physical Sciences (0.084) (0.110) % % Province/region Atlantic (0.065) (0.059) % 0.18* 40% Quebec (0.058) (0.048) 0.07* 47% 0.11* 63% Manitoba (0.094) (0.097) 0.28* 120% 0.15* 52% Saskatchewan (0.090) (0.084) 0.18* 39% % Alberta (0.056) (0.057) 0.20* 52% 0.15* 52% British Columbia (0.066) (0.061) 0.25* 98% % graduates borrowed or was borrowing constrained largely by program rules? Second, which groups of borrowers, if any, were most heavily burdened by their debts? With regard to the first question, we found that the probability of borrowing from the student loan programs varied substantially with field of study, but not in a manner related to expected future earnings. Men were only slightly more likely to borrow than women. We therefore reject the hypothesis that demand factors played a major role in the determination of borrowing patterns and accept the notion that most borrowing was supply-constrained. Regarding repayment, we calculate that the fraction of Bachelors level graduates who reported problems repaying their student loans was, overall, quite small, falling in the 7 8 percent range. Among both men and women, graduates with low current earnings and those in fields likely to have low lifetime earnings reported significantly greater problems with repayment. Holding other variables constant, women reported more difficulty in repayment than men. We found that the proportion repaid within two years of graduation varied systematically with current earnings, the amount owed, and field of study. We estimated separate models for men and women and found that the estimated coefficients of the two models were significantly different. In particular, women repaid their loans roughly at the same rate as men despite their lower earnings. Overall, it would seem that women borrowed only slightly less than men, repaid as quickly as men (despite lower earnings), but reported having significantly more difficulty in repayment. The NGS databases are a rich source of information regarding the experiences of Canadian post-secondary graduates in general, and student borrowing in particular. At the same time, there are many interesting and important questions that we have not been able to answer. First, we cannot determine how many potential students did not go to college or university because they feared the burden of the loans they would have to take out in order to attend. Given the increase in borrowing levels in the 1990s, this question is more pressing now than it was when our 1990 graduates began their postsecondary careers. A new Statistics Canada time series survey, called the Youth in Transition survey, will allow this question to be addressed in the future. Second, we were not able to analyze repayment problems using any objective measure of debt burden, relying instead on self-reported difficulties. Student loans are part of the increasing consumer debt loads that have led to record rates of personal bankruptcy in Canada and the lack of information on other debts in the NGS limits our analysis in this regard.

15 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) Appendix A. Past studies of student loan debt burden To study student loan debt burdens, researchers have tried two general lines of attack. 26 First, researchers used survey data either to calculate rough indicators of debt burden or to ask debtors directly whether or not their loans were causing financial hardship. Other researchers used survey data to approach the issue indirectly by looking at various aspects of the economic situation of student loan borrowers (such as their access to credit or their propensity to purchase major consumer durables). Second, administrative data on the incidence of student loan default and of bankruptcies involving student loans have been analyzed on the assumption that default and bankruptcy indicate financial hardship. A.1. Measuring hardship using survey data A first group of studies analyzed survey data that measured the individual circumstances of particular borrowers. One way to measure hardship in the survey data is simply to ask borrowers if they are experiencing financial hardship as they repay their loans. For example, Baum and Schwartz (1988) surveyed approximately 2000 American student loan borrowers who were both in repayment and whose loans were guaranteed by the Massachusetts Higher Education Assistance Corporation. Their analysis was specifically aimed at investigating the extent to which educational debt is burdensome for borrowers (p. 3). Baum and Schwartz asked several questions designed to elicit respondents subjective impressions of the burden of loan repayment. They summarized the results of those questions as follows (p. 7): Based on the responses, we estimate that approximately 30% [of the respondents] perceive significant hardship resulting from their loan repayment. About one-half of the respondents do not perceive measurable hardship. Ten years later, Baum and Saunders (1998) surveyed a similar population. The level of borrowing had increased fairly dramatically. The average debt level, in current dollars, was $18,600 in 1997 as compared to $7500 in The perception of burden had also increased. For example, 27 percent of the 1987 respondents reported that they had experienced more hardship than anticipated; the corresponding 1997 number was 36 percent. A potentially important change was in the level of non-education debt held by student loan borrowers. In the 1991 version of the survey (Pedalino, Chopick, Saunders, & McHugh, 1991), 10 percent of respondents had non-education debt greater than $1000 per month. In the 1997 survey, 35 percent had non-education debt of more than $1000. Much of this debt may be mortgage debt but, according to Blair (1997), average credit card debt among undergraduate students was about $2200; for graduate students, the average was $ These levels of non-education debt underscore the inadequacy of debt-to-income standards that are based only on student loan debt. Using the Canadian NGS, Finnie and Schwartz (1996, pp )) used ratios of total student debt to earnings to analyze the burden of student loan repayment for Canadian borrowers who had graduated from a post-secondary institution. Such debt earnings ratios will understate the true debt burden because of the omission of other debts and because those who were not working were omitted (since they had no earnings and thus had an undefined ratio of debt to earnings). Even so, for those who received bachelor s degrees in 1990, the median debt/earnings ratio was 0.28 for men and 0.32 for women. The ratios for other degree types and other cohorts were somewhat lower. As a rough indicator of the substantive magnitude of the Finnie and Schwartz ratios, note that Sullivan, Warren, and Westbrook (1989) estimated that the median ratio of total non-mortgage debt to income for their sample of American bankrupts was 0.7. Baum and Saunders (1998) report that the median ratio of total student loan debt to annual gross household income in their American sample was 0.5. The burden of student loans, however, is probably lower than the burden of an equivalent amount of non-mortgage debt because student loans are paid off over ten years (rather than the two or three years that is typical of non-mortgage debt) and thus requires smaller monthly payments. 28 In addition, the above-mentioned Canadian Interest Relief program provides support to those experiencing demonstrable difficulty. Some have argued that if former students face too high a debt upon leaving school, they will be forced to forego careers that are not remunerative, to postpone marriage, 26 An early group of studies, not discussed here, estimated the proportion of a typical starting salary that might reasonably be spent on debt repayment and then compared that proportion to the amount required to repay debts of various amounts. See, for example, Daniere (1969). 27 These numbers are reported in Baum and Saunders (1998), p The empirical literature on bankruptcy generally suggests (though without much theoretical or empirical support) that a debt/earnings ratio of one (that is, a debt load equal to one year s earnings) indicates that bankrupts cannot pay their debts.

16 512 S. Schwartz, R. Finnie / Economics of Education Review 21 (2002) to postpone having children, or in general, to be unable to participate fully in adult life. Baum and Schwartz (1988) asked their survey respondents if their loan repayment obligations had delayed home ownership, car ownership or living apart from parents. At that time, these proportions that said they had delayed these activities were fairly small 23, 17 and 13 percent, respectively. The same questions were asked in the Baum and Saunders survey in 1997 and the percentages reporting delays had increased to 40, 31 and 21 percent. A.2. Measuring hardship using administrative data Low-income students are high-risk borrowers with limited labour market experience and limited access to collateral. Private institutions are usually unwilling to make unsecured loans to such borrowers. That reluctance, combined with the government s desire to make post-secondary education accessible to all, regardless of family income, was the rationale for the provision of government loan guarantees. Most borrowers repaid their loans; the loans of the minority who did not repay fell into default. As late as 1980, student loan default was not perceived as a great problem in some quarters. At that point, about 9 percent of CSLP borrowers had defaulted on their loans. 29 By 1990, however, the proportion of borrowers who defaulted on their Canada Student Loan had climbed to one in six and the Auditor-General of Canada (Auditor- General of Canada, 1990, pp ) was quite concerned both about that rate and about the government s efforts to reduce it. Combining estimates of the rate of default and the rate of repayment after default, Finnie and Schwartz (1996) estimated that between 10 and 12 percent of the dollar value of student loans was lost through default. The role of student loans in personal bankruptcy is even less well-documented. When Brighton and Connidis (1984) studied those seeking bankruptcy protection in 1977, student loan debt was not an important element among the debts of those in their study sample. The increase in borrowing over time, however, had an impact on this extreme form of debt burden. According to the CSLP, the cost to the federal government for student loans in bankruptcy, has vaulted from $30 million in to $70 million in References Auditor-General of Canada, Canada. (1990). Report of the Auditor-General of Canada to the House of Commons. Ottawa. Baum, S., & Saunders, D. (1998). Life after debt: Results of the National Student Loan Survey. Mimeo, Boston: The New England Education Loan Marketing Corporation. Baum, S., & Schwartz, S. (1988). The impact of student loans on borrowers: Consumption patterns and attitudes toward repayment. Mimeo, Boston: Tufts University. Blair, A. (1997). The changing credit environment. Boston: The New England Education Loan Marketing Corporation. Brighton, J., & Connidis, J. (1984). Consumer bankrupts in Canada. Ottawa: Consumer and Corporate Affairs Canada. Cameron, N. (1984). Money, financial markets, and economic activity. Toronto: Addison-Wesley. Council of Ministers of Education, Canada. (1980). Report of the Federal-Provincial Task Force on Student Assistance. Ottawa. Daniere, A. (1969). The benefits and costs of alternative federal programs of financial aid to college students. In The economics and financing of higher education in the United States: A compendium of papers submitted to the Joint Economic Committee. Washington: US Government Printing Office. Finnie, R., & Schwartz, S. (1996). Student loans in Canada: Past, present and future. Toronto: C.D. Howe Institute. Finnie, R., & Garneau, G. (1996a). An analysis of student borrowing for post-secondary education. Canadian Business Economics, 4, Finnie, R., & Garneau, G. (1996b). Student borrowing for postsecondary education. Educational Quarterly Review, 3. Hansen, J. (1991). The shifting roles of parents and students. In J. P. Merisotis (Ed.), The changing dimensions of student aid. San Francisco: Jossey-bass. Little, D. (1997). Financing universities: Why are students paying more? Education Quarterly Review, 4 (2). Pedalino, M., Chopick, C., Saunders, D., & McHugh, S. (1991). The New England Student Loan Survey II: Final report. Boston: Massachusetts Higher Education Assistance Corporation and The New England Education Loan Marketing Corporation. Schwartz, S. (1999). The dark side of student loans: Debt burden, default and bankruptcy. Osgoode Hall Law Journal, 37 (1&2). Sullivan, T., Warren, E., & Westbrook, J. (1989). As we forgive our debtors. San Francisco: Jossey-Bass. 29 The Report of the Federal-Provincial Task Force on Student Assistance (Council of Ministers of Education, 1980, p. 141) asserted that [t]he amount ultimately requiring write-off to date is about 1% of the value of default claims paid though this 1% number does not seem justified by the statistics included in the report. Nonetheless, the Task Force concluded that the existing default rates on the CSLP and on provincial loans were not unduly out of line and did not point to any particular weakness or problems with the concept or administration of the programs (p. 143). 30 See Human Resources Development Canada (HRDC), Ensuring Opportunities: Access to Post-Secondary Education, online: HRDC postgraph/va3fe.html (date accessed: 11 September, 1999). For more information on the relationship between student loans and bankruptcy in Canada see Schwartz (1999).