Multistate period life tables were used in this report to develop projections of numbers of new Ph.D.s that would be needed in the future to sustain certain growth rates of the labor force. This Appendix describes briefly the method used to generate these results. ^{1}

Life table techniques are sometimes the only way to obtain estimates of certain statistics describing mobility and career characteristics of a population, especially those related to rates of occurrence of events, duration of time spent in an activity, and rates of attrition or exit from a population (due to death, retirement, job changing, etc.) even when we have not observed the full lifetimes of the scientists with our data (which is often the case with most data sets). ^{2} Moreover, life table methods provide a useful way of organizing various age-specific rates (rates of entering the labor force, changing jobs, moving abroad, retiring, and dying) into a logical framework, which can then be used to make projections of various characteristics of a population, such as its age distribution.

Multistate life tables, an extension of basic life tables, allow greater complexity to enter the analysis: people can enter as well as exit a population and can move back and forth across a variety of states within a population. Life tables may be classified as period versus cohort life tables. The Panel on Estimation Procedures decided that it was more practical, for the purposes of making projections, to use the former.

Construction of period life tables involves taking agespecific transition rates (job changing, unemployment, retirement, and death rates) prevailing during a particular period (e.g., 1989-1991) and applying them to a hypothetical (synthetic) cohort of people (actually a “synthetic ” cohort of new Ph.D.s). Probabilities are then calculated of entrance or exit from a state (probability of entering a postdoctoral position, for example) and length of time spent in various states, as implied by the life table. Statistics of interest were developed for three time periods (1985-1991, 1979-1985, 1973-1979). Period life table results are conventionally referred to as “expected” quantities (i.e., “expected fraction of people who...,”, “expected length of time..”) because of the nature of the methodology: constructing a single hypothetical Ph.D. cohort that experiences the current transition rates taken from a variety of Ph.D. cohorts.

The data for the life table analysis come from the longitudinal Survey of Doctorate Recipients (SDR), a sample survey that follows a group of Ph.D.s over time, interviewing basically the same people every 2 years. A general description follows (for details, see NRC, 1991 SDR Methodological Report, forthcoming). In 1973 an initial sample of science or engineering Ph.D.s living in the United States was drawn, and those sample members have been followed through time. New Ph.D.s enter the SDR in 1975 and each subsequent SDR year (1977, 1979,...1991) and are followed over time as well. Individuals are followed until they reach a certain cutoff point that depends on the survey year at which they entered [typically 42 years after the Ph.D., although in recent SDR waves, they are followed until they reach age 70 or until they drop out for other reasons (non-response or death)].

The form of the data on which the life tables are based consists mostly of large sets of transition tables constructed from the SDR by National Research Council (NRC) staff (but death rates are obtained from TIAA-CREF data from the late 1980s, taken from Bowen and Sosa, 1991). For every pair of biennial survey-interview years (“waves”) in the SDR (1973-1975 as the first pair, 1989-1991 as the last pair), the number of people moving between various states within those 2 years were obtained. These states were: postdoctorate, R&D employment ^{3} within one's broad Ph.D.

Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.

Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 149

APPENDIX G
MULTISTATE LIFE TABLE METHODOLOGY AND PROJECTIONS
Multistate period life tables were used in this report to develop projections of numbers of new Ph.D.s that would be needed in the future to sustain certain growth rates of the labor force. This Appendix describes briefly the method used to generate these results. 1
Life table techniques are sometimes the only way to obtain estimates of certain statistics describing mobility and career characteristics of a population, especially those related to rates of occurrence of events, duration of time spent in an activity, and rates of attrition or exit from a population (due to death, retirement, job changing, etc.) even when we have not observed the full lifetimes of the scientists with our data (which is often the case with most data sets). 2 Moreover, life table methods provide a useful way of organizing various age-specific rates (rates of entering the labor force, changing jobs, moving abroad, retiring, and dying) into a logical framework, which can then be used to make projections of various characteristics of a population, such as its age distribution.
Multistate life tables, an extension of basic life tables, allow greater complexity to enter the analysis: people can enter as well as exit a population and can move back and forth across a variety of states within a population. Life tables may be classified as period versus cohort life tables. The Panel on Estimation Procedures decided that it was more practical, for the purposes of making projections, to use the former.
Construction of period life tables involves taking agespecific transition rates (job changing, unemployment, retirement, and death rates) prevailing during a particular period (e.g., 1989-1991) and applying them to a hypothetical (synthetic) cohort of people (actually a “synthetic ” cohort of new Ph.D.s). Probabilities are then calculated of entrance or exit from a state (probability of entering a postdoctoral position, for example) and length of time spent in various states, as implied by the life table. Statistics of interest were developed for three time periods (1985-1991, 1979-1985, 1973-1979). Period life table results are conventionally referred to as “expected” quantities (i.e., “expected fraction of people who...,”, “expected length of time..”) because of the nature of the methodology: constructing a single hypothetical Ph.D. cohort that experiences the current transition rates taken from a variety of Ph.D. cohorts.
The data for the life table analysis come from the longitudinal Survey of Doctorate Recipients (SDR), a sample survey that follows a group of Ph.D.s over time, interviewing basically the same people every 2 years. A general description follows (for details, see NRC, 1991 SDR Methodological Report, forthcoming). In 1973 an initial sample of science or engineering Ph.D.s living in the United States was drawn, and those sample members have been followed through time. New Ph.D.s enter the SDR in 1975 and each subsequent SDR year (1977, 1979,...1991) and are followed over time as well. Individuals are followed until they reach a certain cutoff point that depends on the survey year at which they entered [typically 42 years after the Ph.D., although in recent SDR waves, they are followed until they reach age 70 or until they drop out for other reasons (non-response or death)].
The form of the data on which the life tables are based consists mostly of large sets of transition tables constructed from the SDR by National Research Council (NRC) staff (but death rates are obtained from TIAA-CREF data from the late 1980s, taken from Bowen and Sosa, 1991). For every pair of biennial survey-interview years (“waves”) in the SDR (1973-1975 as the first pair, 1989-1991 as the last pair), the number of people moving between various states within those 2 years were obtained. These states were: postdoctorate, R&D employment 3 within one's broad Ph.D.

OCR for page 149

field (biomed, behavioral, other), non-R&D employment within broad Ph.D. field, employment outside of broad Ph.D. field, out of labor force or unemployed (combined), leaving the country, retirement, and death. All of the biennial transition proportions were obtained by 2 year age group by broad Ph.D. field, and by sex. The survey observations (i.e., people) in one set of biennial transitions are often the same people in subsequent sets (though older). 4
Another data ingredient for the life tables is the distribution of states (same as above) of “new entrants to the SDR” for SDR waves 1975-1991, again by age, sex, and Ph.D. field. These are used as estimates of numbers of new Ph.D.s in each survey year.
These transition data sets constructed by NRC staff were transformed into proportions to be used as input into a Multistate Life Table program (Tiemeyer and Ulmer, 1991). Initial work involved explorations of data quality, sample sizes, and the stability of rates over time. To have large enough sample sizes for (what we would hope to be) reliable estimates of sex differences in career patterns as well estimates of how the career patterns have changed over time, it was necessary to aggregate the data into three broad time periods (as opposed to looking at a larger number of time periods): 1985-1991, 1979-1985, 1973-1979.
Projection Models
Life table construction begins by calculating a matrix containing the proportion of individuals exiting an origin state for each possible destination state between ages x and x+2 (in our case). This matrix is called Mx.
Our projection models hold the population of those employed “in field” to some constant growth rate. The following algorithm is used:
Survive the current specified Ph.D. population forward 2 years.
Calculate the number of individuals employed “in field”.
Calculate the differences between the target “in field” population and the number “in field” in the survived current population. This yields the number of new entrants needed to increase the “in field” population to its target size.
Divide the result of (3) by the proportion of new entrants who enter an “in field” employment state on receiving their Ph.D..
Use the result of (4) as the number of new entrants who would have had to enter the population between year y and y+2 to attain the target “in field” population. Add these individuals into the life table, distributed approximately by age and destination state.
Let Nx,y represent the number of individuals in the specified Ph.D. population in each employment state at age X for a given year Y. Nx,y is a k by k matrix, where k equals the number of states in the model. The columns indicate origin states (in the base year) and the rows destination states. So Nx,1995[4,1] would equal the number of people who were in the 4th state (out of field employment) in 1995 who were in the 1st state (in field post-doc) in 1991. For the base year, the off-diagonal elements of Nx,1991 are all 0 and the ondiagonal elements are equal the number of individuals age X in 1991 in the specified Ph.D. population in each employment state.
Let N-x,y represent the number of individuals in each state (by origin state in 1991) in year y, BEFORE new entrants between year y and y-2 are added into the life table. Then N-x,y is given by:
Let F1991 represent the total number of individuals in the specified Ph.D. population employed “in field” in year 1991. Then F1991 is given by:
where x represents age, o represents origin of state, d represents destination state, Nx,1991[o,d] represents the dth row and the oth column of Nx,1991' and where states 1 through 3 represent the employed “in field” states.
Let F-y represent the total number of individuals in the specified Ph.D. population employed “in field” in year Y who were in the specified Ph.D. population (although not necessarily employed “in field”) in year Y-2. Then F-y is given by:
Let G represent the assumed 2-year growth rate for Fy. Then the target employed “in field” population size for any given year is:
Target “In Field” Population Size (y) = F1991 · (1+G)y·1991
Let Dx represent the proportionate distribution by age and state of new entrants to the specified Ph.D. population over the two year period between Y and Y+2. Dx is a 1 by k vector with each column representing the proportion of all new entrants who are age X who enter that state on receiving their Ph.D. Summing Dx across all ages and states should equal one.
Finally, let R represent the proportion of all new entrants who enter an “in field” state on receiving their Ph.D. Then R is given by:

OCR for page 149

Given Nx,1991, Mx, Dx, and G, then Nx,y can be calculated for any y greater than 1991 (in increments of 2-years) by iterating through the formula:
which expands to:
where I is a k by k identity matrix, S is a k by 1 vector of ones, and the symbol x designates an element-wise matrix multiplication operation. (The operation ((S • D) x I) merely takes the Dx vector and turns it into a matrix with the elements of Dx on the main diagonal and zeros on the off diagonals).
The total number of new entrants is given by the component:
Projections were made separately for each of 4 populations:
biomedical Ph.D.'s in biomedical employment fields,
non-biomedical Ph.D.'s in biomedical employment fields,
behavioral Ph.D.'s in behavioral employment fields, and
non-behavioral Ph.D.'s in behavioral employment fields.
To illustrate the use of life table analysis in generating projections of workforce variables, the Panel, as an exploratory exercise, chose to generate estimates of job openings. Given the uncertainty associated with efforts to project demand, the Panel examined three growth rate scenarios based on the average annual growth in the biomedical and behavioral science workforces between 1981 and 1991: zero growth; one-half the 1981-1991 average annual growth; and the average annual growth. 5 Estimates of “net separations” 6 were generated using the life tables. Estimates of needed job openings for the alternative growth scenarios were derived by adding to these separations the number of additional job openings that would need to be created to attain the particular target rate of growth.
In generating the estimates of job openings, the following assumptions were made:
There is never a negative number of new entrants. If there is a surplus in the employed “in field” population at a given year, no new entrants are added to the life table for that year.
The ratio of behavioral Ph.D.'s to non-behavioral Ph.D.'s employed in behavioral fields remains constant. That is, both Ph.D. populations increase or decrease at the same rate. The same assumption is made for the models of biomedical and non-biomedical Ph.D.'s in biomedical employment.
The age/destination state proportionate distribution, Dx, is taken from the age/destination distribution observed among new entrants between 1985 and 1991.
The age/origin state distribution for the current population, Nx,1991, is calculated by taking the age-specific origin state distribution among the current Ph.D. population between 1985 and 1991 and applying it to the age distribution of the 1991 current population.
The age-specific 2-year transition proportions, Mx, used to survive the current age-distribution is taken from the observed transition proportions between 1985 and 1991.
The committee's Panel on Estimation Procedures will extend this work and will prepare a separate report for release in 1994 on the role of multistate life table methods in the estimation of national need.
NOTES
1. Methodological detail is available on request from NRC/OSEP Studies and Surveys Unit (Memorandum by Peter Tiemeyer, September 30, 1993). A general discussion of multistate life tables can be found in Keyfitz (1985).
2. In our particular project, however, we began with transition rates as the basic input data, and derived other life table statistics from those rates.
3. We define R&D employment to be basic or applied research, management of R&D, or development and design of systems and products; it is based on the individual's self-report of primary or secondary work activity.
4. With respect to the treatment of missing data: in general, to enter into the calculation of a biennial transition table, an individual case was required to have valid survey data on age and Ph.D. field and valid data for both of the survey years (for that transition table) on employment field (biomedical, behavioral, etc.) and employment status (postdoctorate, employed, retired, etc.). We developed decision rules for the treatment of all of these variables to handle various conditions (available on request). For example, work activity (i.e., R&D vs. non-R&D) could be missing if the person's employment field was other than biomedical or behavioral (because one of the “states” of the model is “employed outside of Ph.D. field” and those who are out of labor force, retired, or out of the country could be missing employment field and “work activity ”.

OCR for page 149

5. The 1981-1991 average annual growth rates were: 4.25 percent per year for the biomedical sciences workforce and 3.5 percent per year for the behavioral sciences workforce.
6. Net separations are defined in this analysis as losses arising from death, retirement or outmobility to another state minus gains from inmobility of experienced scientists from other states of employment. Alternative definitions will be explored in subsequent work by the Panel on Estimation Procedures.
REFERENCES
Bowen, W.G. and J.A. Sosa 1989 Prospects for Faculty in the Arts and Sciences. Princeton, NJ: Princeton University Press.
Keyfitz, N. 1985 Applied Mathematical Demography. 2nd Ed. New York: Springer-Verlag.
National Research Council Forthcoming 1991 SDR Methodological Report. Washington, D.C.: National Academy Press.
Tiemeyer, P. and G. Ulmer 1991 MSLT: A Program for the Computation of Multistate Life Tables. Center for Demography Working Paper 91-34, University of Wisconsin.