CHAPTER 3
OBJECTIVES AND APPROACHES OF FORECASTING MODELS
There are many potential users of models that forecast demand and supply of workers for particular occupations. One group of users includes students, their parents, and advisors. Others include universities that need to shape the size and scope of their graduate programs, industry planners who need to anticipate worker shortages in key areas of emerging technology, and government funders of graduate education and research, at both state and federal levels, who need to allocate public funds wisely. What is desired in a forecasting model may vary for each of these groups, in terms of horizon, level of detail, and focus.
Burt Barnow, in his workshop paper, described these different objectives:
What we want from the occupational forecasts depends on their use. For career decisions, either by individuals or by training programs on behalf of the participants, we are primarily interested in the employment and pay situation for specific occupations over a fairly long period. For meeting national priorities, however, the number of individuals in a particular occupation is a means to the ends —it may be possible to substitute workers with the right skills but in other occupations to achieve the same ends.
He pointed out that quality, as measured by ability, skill, or credentials, is an important dimension of market adjustment that is
omitted from most forecasting models. When faced with shortages, employers can hire less well trained workers, but they can also redefine jobs. It is important to be able to predict whether shortages will persist. Is there a shortage because employers do not wish to pay the wages needed to attract workers of a desired level of quality? Have employers explored substitution possibilities fully? In the long run, would the shortage remain? This might be the case for scientists and engineers if, for example, students lacked the background in mathematics needed to undertake a science or engineering career. In this case, the lag until supply adjusted to meet demand might be quite long, and changes in immigration policy might be worth considering.
These questions, however, simply underscore the need for better data and more careful modeling of market adjustment. These models could be used to address several questions. How long are adjustment lags? By how much might wages rise in the absence of supply change? How rapidly can students and experienced workers change fields, thereby increasing supply?
One encouraging development on the data front is the development of O*NET by the Employment and Training Administration of the Department of Labor. O*NET will offer data on skills, abilities, and credentials of workers in various occupations, as well as specific descriptions of work performed by occupation. This degree of detail should assist the construction of behavioral models of occupational demand, but O*NET does not provide data specifically about science and engineering occupations at the doctoral level.
Administrators from academia and industry discussed their use of forecasts. John Armstrong, who is a retired vice president for science and technology at IBM, expressed concern that forecasts predicting a decline in demand (e.g., for physicists or hardware engineers) receive much less publicity than those predicting shortages. Although attention to shortages can speed adjustment, industrial personnel forecasting is typically short term and dictated
by the annual budget cycle. When a longer-term shortage is anticipated (for example, of polymer scientists in the mid-1980s), industry does not wait for shifts in government policy. Rather, large firms directly fund university programs designed to increase the supply of a particular type of scientist or engineer. There is a need, however, for publicly funded programs to facilitate retraining and for additional research on occupational choice by college students. Because private firms cannot capture all the productivity gain from worker retraining, society will tend to invest too little in this, at some social cost. Both retraining and modifying to initial training that facilitate retraining are areas where there is a social interest that goes beyond the interests of individual firms or scientists.
Ronald Ehrenberg and George Walker agreed that as academic administrators they make little use of forecasts of demand and supply for scientists and engineers. Federal research funding is far more important than student demand in determining graduate funding. At selective institutions, as the attractiveness of undergraduate majors waxes and wanes, the quality of undergraduate majors varies, and the more attractive majors are more selective. Aside from the imposition of enrollment ceilings for the most popular fields, no attempt is made on the part of institutions to control quantity. The quality dimension is also important in faculty hiring. Good forecasts would be helpful here since universities do have some flexibility in the timing of new hiring. For example, even in the absence of mandatory retirement, the replacement cycle in faculty hiring posits an increase in retirements in 15 to 20 years. 1 Good forecasts of Ph.D. wages and quality would be helpful at that time, so that new hiring could be spread out over time rather than occur primarily when demand is greatest (and wages are at their highest level). Of course, if all universities maintain a similar
1 |
The forecasts of the late 1980s did not anticipate the elimination of mandatory retirement. |
replacement cycle and anticipate it in a similar way, the forecasts of shortages are unlikely to materialize. Walker also stressed the need for graduate students to master strong and transferable skills in science and mathematics and to be encouraged to take risks. Since long-term employment prospects are not likely to be predictable, students need to keep their options open. Narrow graduate training is counterproductive in this regard.
Another user with a practical need for long-term forecasts is TIAA/CREF and other retirement plans. Good forecasts could help such organizations allocate their internal resources to reflect the changing mix of customers (e.g., retirees, young faculty). Such forecasts would also better permit these organizations to prepare and target educational materials for the increasingly diverse groups that make up employees in nonprofit education and research institutions (e.g., part-time and adjunct faculty as well as the traditional base of university faculty and staff). For such organizations, the need for sensible projections based on timely data is really a bottom-line issue, not just a question of national or individual decision making. The timing and magnitude of hiring and retirement have implications for cash flow and the term structure of their investments.
The Bureau of Labor Statistics has been producing an “occupational outlook” for many years and Neal Rosenthal discussed this effort. The outlook makes projections for 10 years and is widely used for career guidance by high schools and post-secondary institutions. The general projection technique is described in the box that follows.
The BLS evaluates its outlook five and 10 years after publication and analyzes what went wrong.2 Over the years it has altered its methodology in the light of these findings, although there are still persistent methodological problems. For example,
2 |
See for example, Monthly Labor Review, November 1997, pp. x-y. |
The BLS Occupational Outlooks The Bureau of Labor Statistics Occupational Outlooks are based on a series of steps that take the forecaster from projections of demand for final goods to projections of demand for occupational employment. In brief, the forecasts are described below.
|
there is an implicit assumption that some relationships are unchanging over time. Thus a fixed relationship is assumed in each industry between the number of jobs and total person-hours. This is clearly problematic, since very strong evidence exists that this relationship has changed and will continue to change as the fixed costs of employment rise relative to variable costs and as the relative importance of overtime cost declines. Also worrisome is the assumption that relationships change at the same rate (linear or exponential) as they have in the past. Finally, the BLS outlook neglects many dimensions in which adjustment may occur, including training and retraining, and especially in response to changes in wages. None of the past changes in the relationships is assumed to
have been affected by anything behavioral—everything is summarized in the time trend. This does not invalidate the BLS framework as a source of information on structural factors that are likely to drive the future market. However, the limitations of the BLS approach need to be communicated to help users understand that behavioral adjustments have not been included. No response is built into time trends in relative occupational wages on either the demand side (where employers substitute capital for labor when relative wages rise) or the supply side (where students move toward occupations in which relative wages are rising).
Although the BLS techniques can be criticized on methodological grounds, they do provide comprehensive occupational forecasts that are in the public domain, although not for doctoral scientists and engineers. On the other hand, both employers and students can respond to the BLS forecast, making it less likely that the predictions will materialize. Moreover, the omission of behavioral responses makes the BLS outlook unreliable as a basis for decisions on federal funding designed to respond to anticipated shortages.
Another user group is the U.S. Congress. It does not need forecasts on a regular basis, but does need them when an issue arises like the adequacy of the supply of information technology workers. In such situations, forecasts are often produced by groups with a vested interest in the outcome of legislation and a limited technical understanding of the rigors of sampling, forecasting models, and labor market definitions. 3 Even if it is difficult to construct supply and demand models in a legislative time frame, a cadre of analysts who have studied the market and can critique the forecasts of special interest groups is valuable. It would be helpful
3 |
In the information technology case, an industry group produced projections based on vacancy rates (ITAA, 1998). In an industry that experiences normally high rates of labor turnover, vacancy rates will be high and will overstate the extent of shortages for the industry. |
if the conclusions made by forecasting models were presented with all of the uncertainties up front. Politicians will pick and choose the results that serve their desired objectives, but at least they will know the caveats. As Skip Stiles, a staff member for the House Science Committee put it, “There needs to be some way of confusing the issues with the facts.” Congress would not only like better forecasts, it would like ways of evaluating the national need for Ph.D. production in science and engineering that is fed by federal funding for research and education. Some of the increases in Ph.D. production recently have come from relatively new doctoral programs. How does that relate to how well science and engineering are being done and implicitly how well taxpayer dollars are being spent? More generally, a better understanding of the higher education system and the incentives that drive it are needed. Modeling should anticipate problems, not just explain past events. Some current trends that are not accounted for in the current generation of supply and demand models are globalization of research, distance learning, education by industry that bypasses traditional institutions of higher education, and the growth of industry/university partnerships.