Initial Market Research Steps Toward a Model of the Military Enlistment Decision

Shel Feldman, Associates for Research in Behavior, Inc.
A. J. Martin, Office of the Secretary of Defense
ABSTRACT - A large-scale tracking study of the propensity of young men and women to enlist in the National Guard or Reserves provided an opportunity to develop and test a regression-based model of a critical life decision. An Ordinary Least Squares (OLS) model comprising nine independent variables replicated in double cross-validations across-randomly split subsamples in each of three independent samples of 1500 persons, in the baseline year. In each sample (non-veteran men, non-veteran women, veteran men), moreover, the model accounted for half the variance of the criterion, based on eight measures of military enlistment propensity. Follow-up interviews of 711 non-veteran respondents, a year later, demonstrated the predictive validity of the measure.
[ to cite ]:
Shel Feldman and A. J. Martin (1981) ,"Initial Market Research Steps Toward a Model of the Military Enlistment Decision", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 329-331.

Advances in Consumer Research Volume 8, 1981      Pages 329-331

INITIAL MARKET RESEARCH STEPS TOWARD A MODEL OF THE MILITARY ENLISTMENT DECISION

Shel Feldman, Associates for Research in Behavior, Inc.

A. J. Martin, Office of the Secretary of Defense

ABSTRACT -

A large-scale tracking study of the propensity of young men and women to enlist in the National Guard or Reserves provided an opportunity to develop and test a regression-based model of a critical life decision. An Ordinary Least Squares (OLS) model comprising nine independent variables replicated in double cross-validations across-randomly split subsamples in each of three independent samples of 1500 persons, in the baseline year. In each sample (non-veteran men, non-veteran women, veteran men), moreover, the model accounted for half the variance of the criterion, based on eight measures of military enlistment propensity. Follow-up interviews of 711 non-veteran respondents, a year later, demonstrated the predictive validity of the measure.

PROBLEM

It often seems difficult to model important life decisions along the lines of problem-solving and consumer purchasing. Although choice of a career and choice of a mate are usually acknowledged to contain elements of rationality, for example, stress is often laid to chance factors and the unique occurrences that differentiate between outcomes. In part, this difficulty may reflect our need to assert uniqueness and free will in those areas most central to our self-concept. In part, it may also arise from a focus on the particular job or spouse chosen -- rather than on the class of jobs or of spouses --and the consequent inability to obtain and analyze data on a large set of similar decisions. A large-scale study on the military enlistment propensity of young men and women offered the opportunity to develop and test a rational model of that important life decision.

BACKGROUND

Successful manning of the All-Volunteer Force requires encouraging more than 650,000 young people to choose military service each year. The voluntary nature of the transaction with potential enlistees has focussed the Department of Defense on the need to consider the nature of its product -- military service --, its positioning, and its promotion efforts through advertising and recruiting, as well as on its pricing variable (enlistment standards). In Fiscal Year 1981, the budget for recruiting will reach $970 million. Much market research will be aimed at understanding the perceptions of potential enlistees, and their decision process regarding military service, in order to carry out recruiting most effectively.

The data reported here are derived from a tracking study of awareness and attitudes toward enlistment into the Reserve Components (Associates for Research in Behavior, 1979). The National Guard and Reserves are part-time military units requiring a specified number of training periods each year, following initial basic training and specialty schools, and provide back-up capability for the Active (full-time) Forces. These components will require nearly 250,000 enlistments in Fiscal Year 1981, from among young men and women with no prior military service (NPS) and those who are veterans of the Active Forces.

The baseline data for this tracking study were collected in Fall, 1978, from a total of 4503 persons. One sample, of 1500, was composed of NPS males, between the ages of 17 1/2 and 26, who had completed no more than two years of college education. A second sample, of 1502, was composed of comparable NPS females. The third sample was composed of 150! male veterans, separated from active service within the previous five years.

The data were collected by phone, in interviews of approximately 30 minutes duration, by trained professional interviewers. The NPS samples were obtained by random digit dialing within specified area codes and exchanges, chosen to form an interpenetrating block design weighted by accessibility to U.S. Army Reserve centers. The veterans were a stratified random sample of all men separated from the Active Forces in the designated time period and eligible for re-enlistment. Each sample was randomly divided into two equal subsamples to test the reliability of various results, including factor analyses and regressions.

Propensity to enlist in the Reserve Components will be reported in two ways. Each is a verbal measure of intention. The "standard" propensity measure indicates the respondent's highest expressed likelihood of serving in any of the six components (Army National Guard, Air Force Reserve, etc.), as indicated on a four-point scale. Thus, if a respondent says she will "definitely" serve in the Navy Reserve, but "probably not" in any of the others, she will be scored "definitely." And if a respondent says he will "probably" serve when asked about each of the components, he will be scored "probably." Responses to questions on each component are highly correlated, particularly for the NPS respondents, who tend not to differentiate between components. In the 1978 data, positive propensity ("definitely" or "probably" enlist responses to at least one component) was shown by 24.7 percent of NPS males, 12.9 percent of NPS females, and 21.0 percent of male veterans.

The "general" propensity measure was developed for use in the regression model, in order to remedy certain deficiencies of the "standard" measure. The "general" measure is a summative index of eight items reflecting military enlistment propensity under a number of conditions. These include the "standard" measure as well as responses to questions on propensity to enlist in the Active Forces, propensity to enlist in the Reserves under different lengths of enlistment contract, and propensity to enlist in the Active Forces or in the Reserves, were there a national service requirement. This summative index has an internal consistency index in the high .80's in each sample, much greater sensitivity than the standard measure, and a far less skewed response surface.

The independent variables on which "general" propensity was regressed were chosen to test a simple decision-making model of the enlistment process. The attempt was made to demonstrate that a single parsimonious model could account for a major share of the variance in each sample. Current work is directed at improving measures and the structural description of the model, given this initial demonstration.

The model posits that enlistment propensity is based on background attitudes, a benefit-cost valuation regarding the prospects of service, and factors peculiar to the particular life situation of the respondent. Background attitudes include general acceptance or rejection of the military in American life, willingness or unwillingness to accept impositions on individual freedom in order to accomplish social goals (in the form of a national service requirement), and degree of need to be with others. Perceived benefits of service in a Reserve Component were measured with respect to opportunities to achieve life goals and opportunities to participate in a team. Perceived costs of service were measured with respect to loss of opportunities to be with family or friends, or to achieve further educational or occupational progress in the time allotted to service, and with respect to other negative outcomes, such as potential harassment by superiors. The respondent's current life situation was measured with respect to job commitment and perceived support for enlistment among current peers.

TABLE 1

"GENERAL" PROPENSITY REGRESSED ON SELECTED VARIABLES

RESULTS

Perceived benefits and costs of service were measured with factorially-derived scales, as were background attitudes toward the military and regarding the need to be with others. The scales were derived from independent principal axis factoring in each subsample of each of the three samples. Comparability was tested using Schonemann's approach (1966; Schonemann & Carroll, 1970). Surviving factors were rotated to maximize interpretability. Resultant scales were constructed, and those with moderate to high internal consistency scores were retained for use in regression analyses of "general" propensity.

The results of the OLS regression analysis in the sample of male veterans are shown in Table 1. These results are clearly consonant with the model proposed. Each of the variables considered contributes significantly to the explanation of "general" propensity, and virtually half of the variance is explained, even without any correction for attenuation. (It should be noted, however, that none of the variables offers great leverage for altering "general" propensity, when evaluated at the point of means.)

The regression results for the male veterans sample were initially developed in each of the two random subsamples involved, and double cross-validated. Moreover, the same model applies to each of the NPS samples, with double cross-validation, accounting for 48.8 percent of the variance in "general" propensity among NPS males, and 47.5 percent among NPS females. With one exception, the variables involved are all significant in each of the other samples as well. (The relative importance of some are sample-dependent, however. For example, the estimated weight of "acceptance of the military" among NPS females is less than half as large as it is among NPS males or veteran males.)

The one exception to the general significance of estimated weights across samples occurs with respect to opportunity costs. The greater the perception of service in the Reserves as taking time from family, friends, job, or school, the lower the enlistment propensity of male veterans. However, this perception is not significantly related to "general" propensity among NPS males, and it is positively related to "general" propensity among NPS females. Presumably, these differences reflect differences in the life cycle position and the commitments of the members of the different samples, and should be explicitly incorporated in the interaction terms of more complete future structural models.

The regression equations were applied to the data from new independent samples in the Fall, 1979 wave of the tracking study. They accounted for between 20 and 30 percent of the variance in the several samples, despite changes in the wording of certain specific items in the predictors, changes in the composition of the "general" propensity measure, and changes in the social and political context (including the onset of the Iranian crisis).

RESULTS OF A FOLLOW-UP STUDY

The major implication of the regression analyses for this paper is that a single parsimonious OLS regression model can capture half the variance in a verbal measure of intention -- intention to undertake or not to undertake an important life change.

But is there any predictive validity to the verbal measure of intention? A follow-up study provided an opportunity to test this.

A year after the original interview, attempts were made to call every NPS respondent in the original sample who had indicated a definite or probable intention to join any of the Reserve Components in Fall, 1978. Attempts were also made to call a random sample of other NPS respondents, of equal size. In total, new interviews were obtained with 372 respondents who had expressed positive enlistment propensity (67 percent of the target group), and with 339 others (60 percent of that target group). These interviews, of about 15 minutes duration, obtained data for estimating the stability of various measures, but concentrated on obtaining reports of important behavior in the past year. The focus of behavioral reports included marriage, job changes, going to school, and military-related incidents, such as responding to military advertising, seeing a recruiter and actually applying to join the military.

Two analyses are of immediate interest. The proportion of respondents with positive enlistment propensity on the "standard" measure who engaged in each of several enlistment-related behaviors is shown in the left-hand column of Table 2. The proportion of others who engaged in each of chose behaviors is shown in the center column. It is evident that the several behaviors form a continuum. (Guttman scalability is .85.) More important for present purposes, positive propensity respondents are clearly more likely to engage in each. In fact, the ratio of the proportion of positive propensity respondents to others increases at each step.

TABLE 2

PREDICTIVE VALIDITY OF PROPENSITY MEASURES

The correlation of 1978 "general" propensity with 1979 reports of engaging in each of the relevant behaviors, over all respondents, is shown in the right-most column of Table 2. Each is statistically significant. The lowest coefficients are found in relation to behaviors that are (compared to applying) largely under the control of others (being accepted for service and then actually joining).

DISCUSSION

Whatever its defects, the verbal measure of propensity to enlist in the military does predict whether or not a respondent is likely to take some steps toward enlistment. Not surprisingly, predictive validity is greater for more proximate behaviors, and for those more under the control of the respondent. This should be kept in mind in further measurement work and further modeling. It may also suggest some shifting of recruiting and advertising strategy away from stress on the final goal of enlistment. Given that a sequence of behaviors is clearly involved, it may be useful to shift more resources to encouraging more immediate behaviors, such as seeing a recruiter.

It cannot be said, at this time, that propensity is the outcome of the variables included in the OLS regression model. The last two decades of attitudinal research show that attitudinal changes, behavioral changes, and changes in the social context are related in a far more complex fashion than was formerly accepted. Ongoing work is focusing on more complex modeling of the enlistment process, including recognition of simultaneity effects. Nonetheless, it is clearly possible to develop a summary index of enlistment propensity from measures of other attitudes, and to predict the likelihood of enlistment-related behavior from scores on that index. Further development of a rational model of this important life decision seems appropriate and useful.

REFERENCES

Associates for Research in Behavior (1979), A Tracking Study Regarding Issues Related to Recruitment of Enlisted Personnel for the Reserve Components, Philadelphia, Pennsylvania.

Schonemann, P. H. (1966), "A Generalized Solution of the Orthogonal Procrustes Problem," Psychometrika, 31 (March), 1-10.

Schonemann, P. H. and Carroll, R. M. (1970), "Fitting One Matrix to Another Under Choice of Central Dilation and a Rigid Motion," Psychometrika, 35 (June), 245-255.

----------------------------------------