Assessing the Effects of Individual Augmentation (IA) on Active Component Navy Enlisted and Officer Retention

Abstract : This report summarizes the results of an analysis of whether individual augmentation (IA) deployment affects retention rates for Navy enlisted personnel and junior officers. The analysis compared retention rates between those personnel who have been deployed via IA to equivalent cohorts of Navy personnel who have not been on an IA deployment. Retention rates were compared in three different ways: aggregate comparisons, comparisons by individual demographic categories, and comparisons based on standard statistical modeling techniques (logistic regression), in order to simultaneously control for all the demographic and other observable characteristics. Overall, the analysis found little evidence that IA deployment is hurting retention rates among those who have experienced one or more IA deployments. In fact, in almost all of the comparisons, the retention rates of those who have had one or more IA deployments were higher than the retention rates of their Navy colleagues who have only been on conventional Navy deployments. The only categories where lower retention rates were definitively identified were for E-3s and E-4s, though the decrease in retention rates was only about one percent.


WHAT DID WE FIND OUT?
Overall, we found little evidence that IA deployment is hurting retention rates among those who have experienced one or more IA deployments. In fact, in almost all of our comparisons, the retention rates of those who have had one or more IA deployments ("IAers") were higher than the retention rates of their Navy colleagues who have only been on conventional Navy deployments ("non-IAers"). See Figures 1 and 2 for aggregate comparison results.  ii The only categories where we found lower retention rates for IAers compared to non-IAers were for E-3s and E-4s and, in these cases, the decrease in retention rates was only about one percent (see Figure 3). 1 Figure 3. Comparison of the percent retained by pay grade and IA status.
E-2, E-4, E-5, and E-9 personnel on IAs had slightly lower retention rates than non-IAers in those pay grades.
Given that retention rates for Navy enlisted personnel and junior officers are generally higher for those who deployed via IA, we conclude the following: It is unlikely that IA deployment causes a significant decrease in retention propensity, at least in terms of the personnel outcomes observed thus far.

SOME CAVEATS FOR OUR FINDINGS
We temper these findings with a number of caveats: • Though IA deployments have been occurring for six years now, we were only able to observe retention decisions on a fraction of those who have been on an IA deployment and these were more likely to be individuals 1 We also did find decreases for E-2s and E-9s, but the number of IAers in those groups was too small to be considered definitive.
iii iv who deployed early in Operation Iraqi Freedom (OIF). Hence, the results observed thus far may not be typical of what is yet to come. See Chapter 4 for additional discussion.

•
We were not able to identify those who volunteered for an IA deployment from those who did not. Thus, it is possible that a higher retention rate for volunteers is masking a lower rate for nonvolunteers. See Chapter 4 for additional discussion.
• Similarly, because this is observational data with strong self-selection effects likely present (at least for the volunteers), it is not possible to conclude that there is any causal relationship between IA deployments and increased retention rates.

RECOMMENDATIONS FOR FUTURE RESEARCH
Given the above caveats, we suggest that additional, on-going research is warranted. Some of our recommendations for such research are briefly summarized here.
These and other recommendations are discussed in more detail in Chapter 4.  shorter notice than conventional Navy deployments and the individual is often deployed to a non-Navy unit. For these reasons and others, Navy leadership is interested in assessing whether IA deployments are affecting retention.

A. DESCRIBING INDIVIDUAL AUGMENTEES
As shown in Figure

B. ORGANIZATION OF THE REPORT
In addition to this introductory chapter, this report is divided into three additional chapters. Chapter 2 describes our analytical approach, including the data we used and how we determined when an individual made the decision to stay in or leave the Navy.
Chapter 3 then presents our quantitative results, including both simple univariate comparisons and more complicated multivariate models. Finally, Chapter 4 summarizes our findings, discusses some of the limitations of the study, and provides recommendations for future research. These latter recommendations should be of interest to researchers and policy makers in the office of the Deputy Chief of Naval Operations (Manpower, Personnel, Training, and Education).

CHAPTER 2: ANALYTICAL APPROACH
The analytical approach we chose was to compare retention rates between those personnel who had been deployed via individual augmentation to equivalent cohorts of Navy personnel who had not been on an IA deployment. "Equivalent" means matching by (or controlling for in multivariate models) observable characteristics such as deployment experience, rank/pay grade, warfare specialty/rating, Armed Forces Qualification Test (AFQT) for enlisted personnel, family status, gender, and race/ethnicity.
The goal was to compare cohorts of sailors and junior officers who were both "at risk" of going on an IA deployment and of leaving the Navy. In particular, for enlisted personnel we had to observe at least one decision to either stay in or leave the Navy between March 2002 and September 2008. For junior officers, their initial service obligation had to expire after March 2002 and within a period of time such that we could determine whether they had decided to remain on active duty or leave the service. Paisant (2008) fully describes the logic for the junior officer analysis, so in the following paragraph we describe it for the enlisted personnel.
As Figure 10 illustrates, we divided the enlisted population up into "IAers" and "non-IAers." For both groups, we had to observe at least one decision to either stay in or leave the Navy between March 2002 and September 2008. "IAers" were then defined as someone who had been on an IA deployment and subsequently made a decision to stay in or leave the Navy. "Non-IAers" were defined as those individuals for whom we had observed a decision point, but they either had not ever been on an IA or their decision was made prior to their IA deployment. This latter case is important since at the point where a sailor had made a retention decision he or she had not experienced an IA deployment and hence was a non-IAer at that time. Scheme for including personnel in the analysis and classifying them as IAer or non-IAer.
Implicit in this approach is that we had to ignore individuals for which we did not observe a retention decision. For the non-IAers, as Section A will describe, this left us with hundreds of thousands of observations against which to compare the IAers.
However, there were significantly fewer decision points observed for IAers. This is both due to the small number of IAers compared to non-IAers, but also because more time must expire in order to observe decision points for those who have been on a recent IA deployment. This has implications for future research that we will discuss in Chapter 3.
For each individual, we then compared by, or controlled for, various observable characteristics. As shown in Figure 11, we chose variable characteristics (such as pay grade or family status) a year prior to the decision point, where the logic was that individuals start to form their decision sometime prior to the actual decision point.
Ultimately, we then compared retention rates between the IAers and non-IAers in the aggregate, by subgroups based on demographic characteristics (such as pay grade or family status), and then in multivariate models where we simultaneously controlled for all the demographic and other observable characteristics.
Examples of classification of included personnel as IAer or non-IAer.

A. THE DATA
The data used to model both the enlisted force and junior officers consisted of a To create the analytical data file for the enlisted analyses, we first deleted all the officer records. Many of the SSNs with duplicate records seemed to refer to enlisted personnel who were later commissioned; we removed these (and all records for any SSN with duplicated records in the dataset). We then removed records for those personnel for whom we did not observe a reenlistment decision after March 2002, who did not have any deployment experience, who separated from the Navy involuntarily, or who did not have any data one year prior to their reenlistment decision. As Figure 12 shows, the been on an IA deployment and for whom we were subsequently able to observe a decision to stay in or leave the Navy.

B. INFERRING DECSION POINTS AND DEPLOYMENT EXPERIENCE
We had to infer a couple of important quantities to conduct our analysis, namely the decision point and whether an individual had (non-IA) deployment experience. Here we discuss how we conducted this inference for the enlisted personnel analysis. For the junior officer analysis, please see Paisant (2008).

Defining the Decision Point
The DMDC data contained a variable that indicated the ETS for each individual for each month of data. For any given month, this variable generally contained the number of months remaining in an individual's enlistment contract. However, once the longitudinal data set was assembled, we were able to determine that this variable did not As a result, we used a number of rules to determine if and when individuals reenlisted. These rules were: • If the ETS went to zero and stayed there for the remainder of the data, we determined that the individual left the Navy at the point the ETS hit zero.
• If the ETS went to zero in some month, but became nonzero again after more than six months, we determined that the individual left the Navy at the point the ETS hit zero, and rejoined in the later month.
• If the ETS went from a number greater than 3 directly to zero in some month and became nonzero within six or fewer months, we determined that the data was in error and that no event had taken place.
• If the ETS went from a number less than or equal to 3 directly to zero in some month and became nonzero within six or fewer months, we determined that a reenlistment decision had taken place in that first nonzero month subsequent to the drop.
• If the ETS went from a number greater than 3 directly to zero and was never again nonzero, we determined that the individual had separated involuntarily.
• We recorded a reenlistment (or enlistment) decision in any month in which the ETS exceeded the previous month's ETS by more than 20, except that if such a jump occurred within the first 12 months of the first appearance of the individual in the data, the enlistment was marked at that individual's first month of nonzero ETS, not at the spot of that jump.

Defining Deployment
In order to assess whether an individual had non-IA deployment experience, we used inferred measures in the Proxy Perstempo data. In particular, we relied on the PERS Tempo Subgroup field: see Section 8 of Appendix B of the Proxy Perstempo Codebook (available from DMDC).

CHAPTER 3: RESULTS
We compared retention rates in three different ways: aggregate comparisons, comparisons by individual demographic categories, and then using models to simultaneously control for demographic and other observable characteristics. We present these results in detail for enlisted personnel and summarize the results for junior officers.
More detail for junior officer comparisons can be found in Paisant (2008).

A. AGGREGATE RESULTS
We begin by simply comparing the retention rates between IAers and non-IAers, both enlisted personnel and junior officers. As shown in Figure 13,   The tables in Figure 15 show the raw numbers (junior officers on the left and enlisted personnel on the right).  (left table) and enlisted personnel (right table) who were retained or not by whether they went on an IA deployment or not.
A way to think about these results is in terms of "odds of retention" for each group, which is the fraction retained for that group divided by the fraction not retained.
For the enlisted personnel, the odds that an IAer is retained is 2.01 (i.e., twice as many enlisted IAers are retained as lost), while the odds that a non-IAer is retained is 1.55. In this comparison, higher odds are better. Similarly, the odds of retention for a junior officer IAer is 1.94, while the odds for non-IAer junior officers is only 0.76. Odds of less than one means that more non-IAer officers are lost than retained, as we see in Figure 14.
We can further compare between IAers and non-IAers in terms of an "odds ratio," or the ratio of the odds IAers are retained to the odds non-IAers are retained. This reduces the comparison to one number. For the enlisted personnel, the odds ratio is 1.30 14 15 and for the junior officers it is 2.56. An odds ratio greater than one means that the odds that IAers are retained is greater than the odds that non-IAers are retained. While the odds ratios are a rather complicated way to distill the results of Figures 13 and 14 down into single numbers, we mention them here as they will be useful in Section C to compare these aggregate results with those from the multivariate models.
Regardless of the metric used, it is clear that the aggregate results show IAers have a higher retention rate than non-IAers for both enlisted personnel and junior officers. Of course, these aggregate results may mask retention issues for certain subgroups, an issue we explore in Section B.

B. UNIVARIATE COMPARISON RESULTS
In this section, we evaluate how retention varies between IAers and non-IAers by various demographics: gender, family status, race/ethnicity, and pay grade. The question is whether there is evidence that IAers of a particular demographic have lower retention rates than their non-IA counterparts. We begin with enlisted personnel. In Figure 18, we see that this result continues to hold when we compare IAers to non-IAers by race/ethnicity. However, in Figure 19, we see that the retention rates for some pay grades are lower for IAers compared to their non-IAer counterparts. In particular, we see that the rates are lower for E-2s, E-4s, E-5s, and E-9s. In terms of the    E-2s and E-9s, the number of IAers is too small to reach any definitive conclusion from this comparison: there were only 9 E-2s and 13 E-9s for which we observed a retention decision. However, for E-4s and E-5s, we observed hundreds of retention decisions (specifically, 373 for E-4s and 604 for E-5s), though the difference in retention rates was only about one percent in each case.

Enlisted Personnel
A closer inspection of Figure 19 suggests that there may be a relationship between pay grade and retention, where the difference in retention rates increases with increasing pay grade. To assess this, taking into account the number of individuals observed in each pay grade, we conducted a weighted regression of the difference in percent retained (i.e., percent IAers retained minus percent non-IAers retained).  We hypothesize that it is not an effect of pay grade per se, but rather some other (unobserved) factor that is correlated with pay grade, such as the fraction of IA volunteers within each pay grade. We discuss this more in Chapter 4.

Junior Officers
The story is very similar for junior officers. Figure 21 shows the retention proportions among IAers and non-IAers by gender for this subgroup (see Paisant [2008] for the details of which officers are included here). In both genders, the IAer retention proportion is higher than that of the non-IAers, with a somewhat larger difference among females. Figure 21. Comparison of the percent retained by gender and IA status. For both genders, the percent retained is higher for those who deployed via IA than for those who did not. For all family types, the percent retained is higher for those who deployed via IA than for those who did not.  Figure 23. Comparison of the percent retained by race and IA status. The percentage retained is higher for those who deployed via IA than for those who did not for every race group; although some of the sample sizes are quite small.

PCT Retained by Race/Ethnicity and IA Status
20 Finally, Figure 24 shows that retention percentages are higher for IAers than non-IAers at every rank. ( Figure 24. Comparison of the percent retained by rank and IA status. The percentage retained is higher for those who deployed via IA than for those who did not for each of the three ranks.

C. MULTIVARIATE MODEL RESULTS
While the foregoing comparisons assess the differences in retention rates between IAers and non-IAers by various demographic categories, these categories are only assessed one at a time. It is possible that the previous results could differ in a comparison that simultaneously incorporates all the demographics.
To conduct such a comparison, we employed a standard statistical modeling technique-logistic regression-to construct our models. For those unfamiliar with logistic regression, the Appendix provides a brief overview of the methodology.

Enlisted Personnel
For the enlisted personnel models, we included covariates in the model to account for known retention rate differences among various demographics (gender, race/ethnicity, and family status), covariates that act as surrogates for personnel quality (AFQT, education), a covariate to account for seniority (pay grade), and a covariate to act as a surrogate for changes in the U.S. economy that may affect overall retention propensity (decision year).
We then ran two separate models, one comparing all IAers to non-IAers and a second one comparing only those IAers deployed to Iraq and Afghanistan to non-IAers.
The results are shown in Tables 1 and 2. As described in the Appendix, exponentiating the coefficient for the IA indicator gives the odds ratio for the retention of IAers versus the retention of non-IAers. We see from Table 1 that the odds ratio for all IAers is and, for

Junior Officers
Here we reproduce the results of Paisant (2008). As shown in

CHAPTER 4: SUMMARY
In this analysis, we evaluated whether retention rates for Navy enlisted personnel and junior officers differ between those personnel who have been deployed via IA and their Navy colleagues who experienced conventional Navy deployments. In our models, we have attempted to control for differences in retention behavior attributable to other personnel demographics-such as rank/pay grade, family status, gender, and race/ethnicity-before evaluating the effect of IA deployment on retention.
Overall, we find little evidence thus far that IA deployment is hurting retention rates among those who have experienced one or more IA deployments. In fact, in almost all of our comparisons, the retention rates of those who have had one or more IA deployments were higher than their Navy colleagues who have only been on conventional Navy deployments. The only categories where we found lower retention rates for IAers compared to non-IAers were for E-3s and E-4s and, in these cases, the decrease in retention rates was only about one percent. (We also found decreases for E-2s and E-9s, but the number of IAers in those groups was too small to be considered definitive.) These findings must be tempered with a number of caveats: • Though IA deployments have been occurring for six years now, we were only able to observe retention decisions on a fraction of those who have been on an IA deployment and these were more likely to be individuals who deployed early in OIF. Hence, the results observed thus far may not be typical of what is yet to come. See paragraph A.1 below for additional discussion.
• We were not able to identify those who volunteered for an IA deployment from those who did not. Thus, it is possible that a higher retention rate for volunteers is masking a lower rate for nonvolunteers. See paragraph A.2 below for additional discussion.
• Similarly, because this is observational data with strong self-selection effects likely present (at least for the volunteers), it is not possible to conclude that there is any causal relationship between IA deployments and increased retention rates. For example, it could be that volunteers are also more likely to stay in the Navy and hence the higher retention rates for IAers are simply due to the choice of the IA volunteers.
Furthermore, it is important to emphasize that our results are about aggregate retention behavior, not individual retention propensities. We expect Navy leadership is most interested in the former where, as we have discussed, there is some utility in knowing that retention rates among IAers (at least as observed thus far) are generally higher. However, the latter is also relevant since it is possible that the IA experience does decrease each individual's retention propensity slightly, but not enough to overcome the inherently higher retention propensities in the self-selected volunteer group. Hence, for example, while we observed higher retention rates for the IAers, it may be that they are not as high as they would have been in the absence of the IA program. Thus, we emphasize that in this research we were not able to assess whether: • any particular individual's propensity to remain on active duty was affected by his or her IA deployment experience, nor • whether the retention propensity of individuals who have not yet been deployed as individual augmentees were affected by the possibility they could be sent on an IA deployment.
That said, based on this research, we conclude the following: • With the exception of some junior enlisted pay grades (E-3s and E-4s), the retention rates for Navy enlisted personnel and junior officers is higher for those who deployed on an IA than for other Navy personnel who experienced conventional Navy deployments.

•
The hypothesis that IA deployment causes a significant decrease in retention propensity is unlikely to be true, at least in terms of the personnel 29 outcomes observed thus far. If it was, we would have expected to see lower retention rates for IAers than for non-IAers.

A. RECOMMENDATIONS FOR FUTURE RESEARCH
Given the previous caveats, and that we were not able to assess some groups, we suggest that additional, on-going research is warranted. In particular, we recommend consideration of the following six areas for future research.

Repeat Analysis Annually
While this research has not found any strong negative effects on retention, it is important to keep in mind that outcomes have been observed on only a small fraction of those who have been on an IA deployment. An outcome for enlisted personnel is the decision to reenlist or leave the Navy and for junior officers it is the decision to continue in the Navy after the initial service obligation or leave the Navy. In both cases, it takes between four and six years to observe such an outcome (either from the start of an enlistment contract for enlisted personnel or from commissioning for junior officers).
Since IAs have only been conducted for the past six years, for most of those who have deployed via IA, their decision to stay or leave the Navy has not been observed. Thus, as outcomes are observed for more sailors and officers, the conclusions of this report could change.
Furthermore, it is important to note that in this data we were more likely to observe outcomes for those who deployed earlier rather than more recently. To the extent that those individuals differed in their Navy career intentions from later individuals who deployed via IA, these results could also change. In addition, the course of the wars in Iraq and Afghanistan has changed substantially over the course of the past six years and will likely to continue to change into the future. To the extent that an IAer's deployed experience affects his or her Navy career intentions, these changes in the course of the wars may affect the observed retention patterns.

Identify and Analyze Nonvolunteers
In this research, we were not able to identify those who volunteered for an IA deployment. Presumably, such individuals are more likely to stay in the Navy. If true, and if volunteers were more likely to be senior enlisted personnel, then the observed association between increasing retention and pay grade for IAers may actually be attributable to volunteer status. Or, perhaps more likely, there exists both an effect due to seniority and volunteer status. In any case, we are not able to identify the volunteer effect due to lack of data.
In addition, a relevant analysis, if nonvolunteers can be identified, is to assess the retention patterns of nonvolunteers. That is, if the assumption that volunteers are more likely to stay in the Navy, and because they volunteered are more likely to positively view their IA deployment experience, then in the current analysis, the volunteers may be masking lower retention rates among the nonvolunteers. That is, if there is a negative effect of IA deployment, it is presumably most likely to be observed among the nonvolunteers.
In discussion with Pers-4, it is our understanding that some data is available for some IAers regarding their volunteer status. Though we were not able to obtain that data for this study, future studies should incorporate it, if possible.

Analyze Reservists
As described in Chapter 1, the majority of IAers are reservists. This analysis only considered AC sailors and junior officers. There is no reason to believe that the effects of IA deployment are the same for RC personnel as for AC personnel, and hence these results should not be extrapolated to RC personnel. Indeed, there are many ways in which the two components differ, and one can rationalize many ways in which an IA deployment might have a more positive or more negative effect on RC personnel (compared to AC personnel).

Analyze Mid-Grade Officers and Warrant Officers
Because outcomes for mid-grade officers (defined as O-3s past their initial service obligation decision point through 0-5s) and warrant officers were not sufficiently observed, they were not analyzed in this study. That is, as described in paragraph A.1 above, decision points for many of sailors and junior officers have not been observed in the six years since IA deployments began. This problem is even greater for mid-grade officers and warrant officers who have made a least an initial commitment to a naval career and whose decision timelines are even more extended.
Simply put, not enough time has expired to observe enough mid-grade and warrant officers with IA deployment experience leaving the service. However, as time progresses, such analyses, if they are desired, will become possible.

Evaluate Using Other Outcomes
In this analysis, we have used retention as the relevant comparison measure between those who have been on an IA deployment and those who have not. In the process of conducting the evaluation, however, we removed those personnel who were involuntarily separated, under the assumption that we were interested in comparing the retention and separation rates among those who chose to stay in or leave the Navy.
However, other measures may be relevant. In particular, if IA deployments are causing increases in involuntary separations (say for mental health reasons), then our analysis would not have been able to detect this and such an increase could also be relevant to the question of how IAs are affecting retention in the Navy. Thus, future studies could assess the types and rates of involuntary separation between IAers and non-IAers.

Conduct Survey and Connect Attitudinal to Outcome Data
In these analyses, we have conducted an analysis of the most concrete measure of whether IAs are affecting retention by looking at actual retention behavior. However, this is an evaluation of aggregate behavior and, as such, it cannot assess whether, even 32 though an individual was retained in the Navy, his or her future propensity to remain on active duty has been increased or decreased in some way by the IA experience.
One way to take a step closer to evaluating this and similar questions is by using a survey to collect attitudinal and other data on those who deployed via IA and then connect the survey data to the outcome data. In so doing, it may be possible to assess whether and how various aspects of the IA deployment experience influenced an individual's decision to stay in or leave the Navy.
The odds are defined as the probability that an individual with a particular set of characteristics will stay in the military, divided by the probability that he or she will not.
The odds can be any number between zero and infinity. Odds of one means that an individual with those characteristics is equally likely to separate as not. Odds greater than one means that such an individual is more likely to stay on active duty, while odds less than one means the individual is more likely to separate.
Through algebraic manipulation, we can explicitly estimate the probability of retention for the i th individual, ˆi p , as a function of the coefficients: β can be interpreted as the odds ratio (OR) when X j is a binary characteristic. The odds ratio is simply the ratio of the odds when X j = 1 to the odds when X j = 0 and is roughly equivalent to the relative risk. If OR = 2, then we interpret this to mean that individuals with characteristic X j = 1 are twice as likely, on average, to stay in the service as those with X j = 0. Such a change might have the effect of changing the odds of staying in from 1 to 1 ("even money") to 1 to 2 (representing a change in probability from 0.5 to 0.67), or it might change the odds of staying in from 100 to 1 to 50 to 1 (representing a change in probability from roughly 0.001 to 0.02).
Because of the nonlinear relationship between the ˆi p and theβ , this model cannot measure the effect of changes in β directly on the values of ˆi p .