4. Methodology

Whereas the early cost-benefit studies used relatively simple research methods, today quite complex and sophisticated statistical and other techniques have been developed. However, the underlying concepts and problems have, for the most part, remained the same. The methodology used in cost-benefit analysis outside the world of education essentially applies in its entirety to educational cost-benefit studies but, in addition, the latter give rise to complex conceptual and computational problems of their own.

It is important that the various problems indicated in this section are seen in context and are not taken to invalidate what is still a widely-used and very useful technique.

According to the traditional method of calculating rates-of-return to investment in education from a detailed cost-benefit analysis (we shall refer later to a revised technique that has attracted considerable attention in recent years), the analysis must commence with a tabulation of all the costs and all the benefits of the expenditure in question.

The computation of educational costs is not a simple matter; it is possible to arrive at a number of different definitions of costs, which may result in contrasting figures (Hough, 1981). Nevertheless, the principles involved in calculating costs in education are not essentially different from those involved in calculating costs elsewhere.

To determine the benefits from education is much more difficult and involves philosophical issues relating to the purposes of education and how to assess whether these are being achieved. Economists have tended to concentrate on the relatively hard evidence that exists in most countries that those people with higher levels of education on average receive higher incomes throughout their working lives than people with lower levels of education. These differences, as measured by data known as age-earnings profiles, appear to be relatively stable and consistent over time. It has therefore seemed reasonable to regard the income-stream differentials, or some proportion of them, as attributable to the education received and it has become conventional to use them to measure the benefits from education. Clearly, to do so is not without problems and leaves a number of questions unanswered but efforts to find alternatives have met with difficulties. One of the most interesting alternatives was the attempt to measure the contribution of education directly by comparing the physical output of educated and less educated workers (Jamison and Lau, 1982).

At the outset it is necessary to decide whether to use the Present Value method or the Internal rate-of-return method. This is a rather technical distinction between the former, which deducts the present value (arrived at via discounting) of costs from the present value of benefits to arrive at a net figure, and the latter, which arrives at the rate of discount which equates the total benefits with the total costs. With the former, the rule is:

"Select all projects where the present value of benefits exceed the present value of costs",

whereas with the latter the rule is:

"Select all projects where the internal rate of return exceeds the chosen rate of discount" (Press and Turvey, 1965).

The latter, the Internal rate-of-return method, avoids the difficult problem of which rate of discount to employ in the calculation, and is commonly used.

In many cases the two approaches will give equivalent answers, although this need not necessarily be the case (Cohn and Geske, 1990).

The principal conceptual and other problems that arise in educational cost-benefit computations are as follows:

(i) Which type of cost-benefit analysis is required? There are four possibilities, as under:

	From perspective of the individual	From perspective of society as a whole
Average over all education received	Average private rate-of-return	Average social rate-of-return
Incremental part of education	Marginal private rate-of-return	Marginal social rate-of-return

The social calculus relates the whole of the costs to society to gross (before deduction of income tax) incomes. The private calculus relates those costs borne by the students and/or their families to net (post-tax) incomes.

Which of these is required will depend on the reasons for carrying out the analysis. The social rate-of-return should be important for educational planning since it gives the returns to society as a whole but the private rate-of return shows the basis on which individual students make their investment decisions. Many studies include more than one type. It is also true that, in a sense, all rate of return calculations can be regarded as "marginal", in that they measure the costs and benefits of a marginal increase in investment in education.

(ii) In principle, all costs (opportunity costs, not just money expenditure) and all benefits should be included but in practice this may not be possible and it may be necessary to settle for some degree of approximation; an example would be the need to give an approximate apportionment of capital depreciation.

(iii) Do workers' earnings differentials accurately reflect differences in their marginal productivities? This point has been termed "the Achilles Heel of rate-of-return analysis" (Blaug, 1970). If they do not, there will be a problem in using them as a proxy for benefits in social rate-of-return calculations. They may, for example, reflect:

"traditional hiring practices and a variety of social conventions about the relative worth of different kinds of labour, not to mention the restrictive practices of trade unions and professional associations" (Blaug, 1970)

This point may also be important in connection with public sector employees, who in many Third World countries comprise large percentages, often 50% or more, of the more highly educated people. Cost-benefit calculations rarely include corrections for market imperfections. Similarly, in connection with private rate-of-return calculations, how to allow for "the non-pecuniary attractions of certain occupations that are accessible only to the highly educated" (Blaug, 1970)? No ready method has been found.

Subsequently, there has developed the "screening hypothesis" which suggests that education does not directly affect productivity at all but simply enables employers to identify workers with different levels of ability, one consequence being that an increase in the supply of educated workers leads to "credentialism" as employers demand higher and higher levels of education.

(iv) In using age-earnings profiles as a proxy for educational benefits, as indicated above, it has become conventional to include an "alpha-coefficient" (sometimes called an "ability adjustment") - although it would seem more appropriate to call it an "education coefficient" (Hough, 1967) - adjustment for the proportion of differences in incomes to be attributed to factors other than education, such as innate ability, personality, favourable home background and social class. Following the work of Denison (1964), in many studies the alpha-coefficient is taken to be two-thirds (i.e., this is the proportion of the income differences attributed to education). However, this may be a considerable approximation: Denison's findings related to the USA, to males only, solely to high-school and college levels of education and accepted the validity of IQ test scores (which have been much disputed elsewhere). Denison's findings have been challenged by other writers - Blaug, for example, suggests that for some groups, including university graduates, the figure of 0.66 may be too low but that for secondary-school leavers in the UK the alpha-coefficient may well be less than 0.66. It would indeed be surprising if the same figure applied to all groups of people in all societies:

"The estimates made of the effect of education alone are based on slender evidence, ignoring major studies, and the standard error of the estimates is likely to be large even if the position is accepted" (Vaizey, 1972).

Psacharopoulos (1975) suggests that for developed countries a figure of 0.7 or 0.8 may be more appropriate but rather little is known regarding an appropriate value of the alpha-coefficient for developing countries.

(v) Available income statistics almost always exclude the value of fringe benefits, which may be important in some occupations. Examples would be the provision of subsidised meals, medical care, or transport to and from work.

(vi) Age-earnings profiles should be based on time-series statistical data, i.e. data collected over the whole of the working life, a period of forty years or more. For obvious reasons, these rarely exist and it is necessary instead to rely on cross-section data, i.e. snapshot evidence of cross-sections of society at one moment in time. Such cross-section data may be unduly affected by short-run cyclical changes in the economy, they ignore future changes in the demand and supply of educated manpower and they fail to capture the effects of trends over time, the major one of which in most countries is the incidence of economic growth. Regarding the latter, Becker (1974) in the USA suggested adding the annual expected increase in real income per head and Ziderman (1977) in the UK "conservatively" added 2 per cent per annum to all incomes, as did Blaug, Layard and Woodhall (1969) for India. The effect of such an adjustment on the final computation is considerable; further, to add a fixed percentage adjustment in this way assumes that income differentials will remain constant over a period of some forty years, which seems very unlikely (Hough, 1987). On the other hand, an advantage with using cross-section data is that it is not necessary to correct for the changing effects of inflation over time. Some time-series data has recently started to become available and Psacharopoulos (1985) found evidence that over time the rate of return to education declined slightly in developing countries but remained relatively stable in developed countries.

(vii) How to translate into monetary terms some elements which it may be difficult to quantify, one example being the benefits from university research, some of which may accrue as a spin-off from the teaching process? Again, some degree of approximation or estimation may be necessary.

(viii) The timing of any costs or benefits, especially the latter, where some of the benefits may accrue many years hence. The principle of Discounted Cash Flow is that benefits in the immediate or near future should
figure much more prominently in the final calculation than benefits much further away (the problem is usually less acute on the costs side). Therefore, values need to be discounted over time in order to be expressed in today's value. However, the choice of discount rate may not be easy and has a significant effect on the calculation (although this problem is avoided where the choice of discount rate is not built into the calculation as in the case of cost-benefit ratios).

(ix) The principle of opportunity costs, notably in connection with how to value the input of time by the student into the learning process, commonly valued via income foregone (following Blaug, 1970, although Vaizey, 1972, disagreed with this approach). But if the process of education is pleasureable, as one must hope that it is for most students most of the time, then are we justified in regarding the time so spent as a cost? A significant point in developing countries is that the student's family will often suffer the loss of his/her income, either monetary income or in terms of practical work done, and that primary school children, particularly girls, are often withdrawn from school because their parents need their services at home. The importance of allowing for income foregone may be seen when it is realised that, when it is included, it frequently exceeds the whole of the direct cost of the education in question.

(x) How to allow for the probability of unemployment, which would affect both the calculation of future income streams and also the opportunity cost of the student's time? In many countries, unemployment statistics show little consistency over time and may in any event be inaccurate; therefore, predictions of future unemployment may be subject to considerable error.

(xi) Problems of data availability: the statistical data required may not be available and it may be necessary to make use of some alternative, which may or may not be a good substitute and may involve some degree of approximation. An example would be when Ziderman (1977) needed data relating to income streams for people educated to GCE A level: the nearest substitute he could find was the salary scale for the Executive class in the Civil Service, for which GCE A-level was the normal entry requirement.

This obviously begs the question of whether people with the same level of education but in other jobs would have had higher or lower incomes.

(xii) "Externalities" or spill-over benefits to persons other than those having received the education in question, notably increased incomes to other people brought about by the higher productivity of the educated person. Attempts to quantify spill-over benefits have proved extremely difficult but Becker (1964) estimates that to include them could lead to the original benefits, and thus the ensuing rates-of-return, being doubled.

(xiii) Woodhall (1973) showed that there are reasons for thinking that the rates-of-return to educating women may be considerably higher, perhaps by two percentage points, than the standard computations would show, on account of such factors as the higher probability that more highly educated women will return to work after child-bearing, that more highly-educated women may face less market discrimination than uneducated or less-educated women, that women's non-market work has positive economic value, and that women arguably enjoy increased psychic income as compared to men educated to similar levels. These factors, together with the fact that women tend to be concentrated in public sector employment, such as teaching or nursing, where the value of earnings as a measure of marginal product was more than usually suspect, combined to suggest that rate-of-return studies typically understated the returns to investment in the education of women. It is noticeable that many cost-benefit analyses use data relating to males only.

(xiv) Various other adjustments may be found necessary in particular cost-benefit calculations, depending on the circumstances. An example would be the cost-benefit analysis by Birch and Calvert (1974) relating to the profitability of becoming a teacher in the UK: they found it necessary to adjust teachers' income streams upwards by one-twelfth (= one-month's extra salary) to allow for the "perk" of extra-long holidays.

(xv) No way has been found to isolate the effects of investment in education from other forms of investment in manpower, such as associated medical care, on-the-job training, and even migration. In the absence of any evidence to the contrary, we have to assume that the return to investment in education does not differ significantly from the return to such other forms of investment in human capital (Blaug, 1970).

(xvi) Rate-of-return analysis tells us whether to invest more or less in a particular direction. But how much more or less? This is a question that rate-of-return analysis can not answer, other than:

"to answer 'a little bit more or less' after which yields will have to be recalculated (Blaug, 1970)

And since the effects of any education investment decision may not be felt for some years hence, to undertake such a recalculation in the short term may be impossible.

Rate of return calculations may be biased upwards or downwards, depending on which of various extraneous points have been allowed for. Professor Blaug gave a "Check List of Biases in Rates of Return", as follows:

	Downward Bias (too low)	Upward Bias (too high)
Private Rates of return	1. Lower rates of return to other types of human capital formation (training, health, etc.)	1. Higher rates of return to other types of human capital formation.
	2. Future consumption benefits(?).	2. Present consumption benefits(?).
	3. Non-pecuniary occupational preference of educated people.
	4. Improved quality of education.
	5. Earnings differentials include first-round spill-overs.
Social	1. As 1 above	1. As 1 above
Rates of Return	2. Future consumption benefits (?).
	3. Non-pecuniary occupational benefits taking the form of fringe -benefits.
	4. As 4 above
	5. Earnings below marginal private product(?).	5. Earnings above marginal private product (?).
	6. Excess demand for labour	6. Over-staffing in public sector
	7. Externalities (first-round and second-round spill-overs).

This section has reviewed a formidable list of conceptual and computational problems and adjustments. The effects of at least some of them might be very substantial, for example, the inclusion of spill-over benefits might double benefits whilst the inclusion of earnings foregone might double costs; rather fortunately, perhaps, these might cancel each other out. However, in practice, most rate of return studies do include earnings forgone but exclude externalties. The effects of other possible adjustments should be less. It is, of course, true that at least some of these points also apply in the case of other approaches to educational planning, notably the manpower planning approach, which, for example, also largely ignores spill-over effects. This needs to be borne in mind when the advantages and disadvantages of cost-benefit analysis are being weighed against those of other approaches.

It is also true that other types of investment (e.g. investment in health care, agricultural development projects) also generate "spill-overs" which are often ignored. There may also be spill-over costs as well as benefits. Recent attempts to estimate the environmental impact of investment projects are one way of attempting to measure spill-over costs of investment projects. The "environmental impact" of education may be both positive (e.g. educating children in environmental awareness) and negative (e.g. if emphasis on academic education generates distaste for technical/vocational programmes and occupations).