Classifying U.S. Army Military Occupational Specialties Using the Occupational Information Network

Abstract

Objectives

To derive job condition scales for future studies of the effects of job conditions on soldier health and job functioning across Army Military Occupation Specialties (MOSs) and Areas of Concentration (AOCs) using Department of Labor (DoL) Occupational Information Network (O*NET) ratings.

Methods

A consolidated administrative dataset was created for the “Army Study to Assess Risk and Resilience in Servicemembers” (Army STARRS) containing all soldiers on active duty between 2004 and 2009. A crosswalk between civilian occupations and MOS/AOCs (created by DoL and the Defense Manpower Data Center) was augmented to assign scores on all 246 O*NET dimensions to each soldier in the dataset. Principal components analysis was used to summarize these dimensions.

Results

Three correlated components explained the majority of O*NET dimension variance: “physical demands” (20.9% of variance), “interpersonal complexity” (17.5%), and “substantive complexity” (15.0%). Although broadly consistent with civilian studies, several discrepancies were found with civilian results reflecting potentially important differences in the structure of job conditions in the Army versus the civilian labor force.

Conclusions

Principal components scores for these scales provide a parsimonious characterization of key job conditions that can be used in future studies of the effects of MOS/AOC job conditions on diverse outcomes.

INTRODUCTION

A long tradition of research in organizational psychology and related disciplines has examined occupational differences in worker’s health, ^1,2 job satisfaction, ^1,3 and work performance. ¹ These studies have documented such things as low rates of job satisfaction among workers with jobs featuring a combination of high demands and low autonomy,^4,5 low work performance among workers exposed to high-intensity artificial lighting,⁶ high rates of hypertension among emergency respondents,^7,8 and high rates of suicide among farmers.⁹

Although few comparable studies have been carried out in the military, research has documented significant differences across military occupations in rates of job satisfaction, ¹⁰ injury, ¹¹ disability, ¹² and suicide. ^13,14 A limitation of these military occupational studies is that the denominator population is too small to generate stable outcome estimates for most individual occupations, whereas broadly defined occupational groups, such as two-digit Department of Defense-wide occupation codes, are not designed to define a meaningful dimension in predicting a diverse range of outcomes. One way to address these limitations is to use a conceptual scheme to classify military occupations into a small number of categories using objective ratings of job conditions based on the Occupational Information Network (O*NET) system.¹⁵ O*NET is a system created by the U.S. Department of Labor (DoL) to standardize reporting of occupational statistics based on objective ratings of diverse job requirement and work characteristics across the labor force.¹⁶ The job requirements (knowledge, skills, abilities, education) and work characteristics (activities, tasks, settings) dimensions used in O*NET were selected from dimensions found important in the occupational analysis literature. ^16–18 O*NET version 15, ¹⁹ which was used in the current report, contains 853 occupational groups rated on 246 different descriptive dimensions.

O*NET has been used to examine effects of job conditions on a variety of outcomes in the civilian labor force. ^20–23 However, the complexity and overlap among the 246 O*NET dimensions make it difficult to distinguish clear effects of particular job characteristics. ¹⁵ Principal components analysis (PCA) and exploratory factor analyses (EFA) have consequently been used to derive composite measures of multivariate O*NET item profiles. ^{20,21,24–26} A number of stable components/factors have been found in these studies. It is not clear, though, whether the same components/factors adequately characterize military occupations, as not all DoD occupation codes have direct equivalents to civilian occupations.

The goal of the current report is to derive O*NET-based summary dimensions of military occupations for use in future research on associations between occupational characteristics and soldier outcomes. To accomplish this, we generated O*NET dimension scores for roughly one-third of Army occupations that do not currently have direct equivalents in the O*NET system. We then carried out PCAs across all Army occupations. Our immediate purpose in generating these scores was to use them in the “Army Study to Assess Risk and Resilience in Servicemembers” (Army STARRS; www.armystarrs.org), an epidemiological and neurobiological study of risk and resilience factors for Army suicides. However, the resulting component scores can be used in the study of other topics. ^20,26–28 To that end, an appendix accompanying this article provides a public use file of derived scores for each Army occupation on each component scale generated in the analyses reported here (available at http://www.hcp.med.harvard.edu/ncs/ftpdir/ONET_Methods_Paper_Appendix_Table_I.xlsx).

METHOD

Sample

We built a consolidated Army STARRS Historical Administrative Data Study (HADS) database combining individual- level data across diverse Army and DoD data systems for all Regular Army, Army National Guard, and Army Reserve soldiers on active duty at any time between January 2004 and December 2009, ²⁹ a total of roughly 1.6 million soldiers. Each soldier was classified by Military Occupation Specialty (MOS; for enlisted soldiers and warrant officers) or Area of Concentration (AOC; for commissioned officers). ³⁰ However, MOS/AOC often changes over time as soldiers gain new competencies and are promoted. Furthermore, soldiers vary in the amount of time they are on active duty. We addressed these two issues by converting the person-level records into person-month records; that is, a series of separate records for each soldier for each month of active duty service over the study period. Each person-month was coded according to the soldier’s MOS/AOC during that particular month. The 1.6 million soldiers in the file had a total of 51.1 million person-months (37.0 million in the Regular Army, 5.3 million in the activated Army Reserve, and 8.8 million in the activated Army National Guard). A stratified (by sex, rank, time in service, deployment status, component, and historical time) random sample of one out of every 400 person-months (a total of 111,195 person-months) was selected to produce a dataset of more manageable size. Use of this person-month sample implicitly weighted each MOS/AOC by its frequency of occurrence in the active duty Army. The sample was subsequently reduced to 100,877 by excluding person-months when soldiers were either in training, in the hospital, or in rehabilitation.

Measures

The O*NET Structure

Although O*NET descriptors are organized into six domains, the descriptive dimensions are confined to three of these six: the “worker characteristics” domain (80 dimensions in the subdomains of abilities, occupational interests, work values, and work styles); the “worker requirements” domain (68 dimensions in the subdomains of basic skill, cross-functional skill, and knowledge requirements); and the “occupational requirements” domain (98 dimensions in the subdomains of generalized work activities and work context). Consistent with civilian studies, ^20,25,26 these were the domains used in the current analysis, as the other three domains either contain information specific to individual occupations rather than applying to all occupations (the “occupation-specific information” domain, where we find such information as the speed of typing required to be a secretary), contain information about work experience requirements (the “experience requirements” domain, where we find such information as apprenticeship requirements), or contain labor market information irrelevant to Army occupations (the “workforce characteristics” domain, where we find occupation-specific information on civilian wage and employment rate trends based on Bureau of Labor Statistics data).

Separate scores (each with a 0–100 range) on the “level” and “importance” of the dimension required for the occupation exist for 161 of the 246 dimensions in the worker characteristics, worker requirements, and occupational requirements domains. Only a single score on either level or importance is provided for the other 85 dimensions. As level and importance scores are highly correlated for dimensions containing both (Pearson r range = 0.58–0.99; interquartile range = 0.88–0.98; median = 0.94), PCA/EFA of the O*NET dimensions in civilian samples have either combined the two scores^25,26 or used only the level score. ²⁰ We combined the two scores using both arithmetic and geometric means. Results reported here are based on arithmetic means, as the component loadings using this method were somewhat stronger than when using geometric means.

Mapping MOS/AOC Codes to O*NET Occupations

The Army classifies soldiers into 645 different MOS/AOCs³⁰ that are recorded in the Defense Manpower Data Center (DMDC) Active Duty Master Personnel File. Monthly information is recorded in this file on changes in MOS/AOC. Both “primary” MOS/AOC (the occupation of the soldier’s most recent training) and “duty” MOS/AOC (the job being performed in the month of observation) are recorded. Although primary and duty MOS/AOC are typically the same, they sometimes differ. For example, soldiers with a primary MOS of cook can have a different duty MOS during deployment because civilian contractors are often used as cooks in combat theaters. We focused on duty MOS/AOC rather than primary MOS/AOC in our analysis, as we wanted to focus on the actual day-to-day tasks carried out by soldiers rather than on the duties for which they were trained.

We used a “crosswalk” created by the developers of O*NET and DMDC to assign O*NET codes to the 474 Army MOS/AOCs considered comparable to civilian O*NET occupational groupings (http://www.onetonline.org/crosswalk/MOC/). This crosswalk takes into account that some MOS/AOC numbers changed over time by matching civilian characteristics to MOS/AOCs for appropriate time periods. The other 171 (i.e., 645 – 474) MOS/AOCs were not included in the crosswalk because there is no civilian equivalent occupation (e.g., a mine sweeper). We addressed this problem in two ways. First, we excluded 35 of the 171 because they involved temporary statuses not related to actual jobs (most involving trainees and patients in hospitals or rehabilitation). Second, we created a composite classification scheme for each of the remaining 136 (i.e., 171 – 35) MOS/AOCs to capture the main requirements of that job and then assigned O*NET ratings using the maximum value for each dimension across the multiple O*NET occupational groupings in the composite. In the case of a mine sweeper, for example, we created a composite that included the O*NET occupational groups of Police SWAT team member, skilled heavy equipment operator, and several other civilian occupational groups that captured the key aspects of the skill set and job conditions of an Army mine sweeper. The DMDC reviewed and approved these ratings. These 136 MOS/AOCs were then combined with the 474 already classified to create a set of 610 MOS/AOCs with a score on each of the 246 O*NET dimensions. These 610 MOS/AOCs formed the basis of the analyses reported below.

Analysis Methods

Correlation matrices were constructed to characterize bivariate associations of the 246 O*NET dimensions in the HADS sample of 100,877 person-months. PCA of these matrices was carried out using the “parallel analysis” simulation method to determine the optimal number of components to retain. ³¹ Parallel analysis extracts the maximum number of components that can be reliably distinguished in a dataset. ³² Once a final number of components was selected, an oblique (promax) rotation was used to improve component interpretation. We used oblique rather than orthogonal rotation based on evidence in the civilian literature that significant correlations exist among O*NET component/factor scores. ^20,25,26 SAS version 9.2 was used for all analyses.

RESULTS

Principal Components Analysis of all O*NET Dimensions

Our first analysis aimed to determine the extent to which we could replicate earlier findings regarding the structure among the 246 O*NET dimensions in civilian research. ²⁵ Parallel analysis suggested that three components could be reliably distinguished in the data. These three together explained 65.0% of total item variance. (Table I) Standardized partial regression coefficients in the promax-rotated solution suggest that these components reflect “physical demands” (e.g., worker ability requirements of physical coordination and manual dexterity; worker skill requirements of maintenance, repair, and control of machines and processes; and work contexts involving hazardous exposures); “interpersonal complexity” of work (e.g., worker characteristic requirements of self-control and concern for others; and occupational requirements of dealing with conflict situations); and “substantive complexity” of work (e.g., worker characteristic requirements of flexibility and originality; worker skill requirements of complex problem solving and judgment; and occupational requirements of thinking creatively, supervising, and providing consultation). The interpersonal complexity and substantive complexity components are correlated r = 0.37, whereas the physical demands component is correlated r = 0.21 with the interpersonal complexity component and r = 0.06 with the substantive complexity component.

TABLE I.

Standardized Partial Regression Coefficients of Illustrative High-Loading O*NET Dimensions Based on Three-Factor Promax Rotated Exploratory Factor Analysis of all 246 Dimensions in the Army STARRS 2004–2009 Administrative Records Person-Month Sample (n = 100,877)^a

	I	II	III
I. Physical Demands
Work Context: Exposure to Contaminants	0.89	0.10	0.02
Work Context: Exposure to Hazardous Conditions	0.84	0.14	0.20
Abilities: Manual Dexterity	0.86	0.07	0.01
Abilities: Control Precision	0.89	−0.02	−0.06
Skills: Troubleshooting	0.89	−0.31	0.26
Skills: Operation Control	0.85	0.07	−0.04
II. Interpersonal Complexity
Work Context: Frequency of Conflict Situations	0.07	0.92	−0.07
Work Context: Deal With Unpleasant People	0.10	0.94	−0.32
Work Style: Stress Tolerance	0.13	0.79	0.11
Work Style: Concern for Others	−0.05	0.87	−0.18
Skills: Negotiation	−0.17	0.74	0.21
Skills: Social Perceptiveness	−0.25	0.78	0.33
III. Substantive Complexity
Skills: Complex Problem Solving	0.06	−0.10	0.90
Skills: Judgment and Decision-Making	0.00	0.22	0.77
Abilities: Flexibility in Using Rules, Combining and Grouping Things	−0.03	0.06	0.83
Abilities: Fluency of Ideas	−0.11	0.30	0.72
Work Activities: Thinking Creatively	0.07	0.09	0.75
Work Activities: Analyzing Data or Information	0.01	0.14	0.75

^aBased on exploratory principal axis factor analysis with promax rotation of 246 O*NET dimensions. The number of factors to extract was determined by simulation. All boldfaced coefficients are significant p < .001. Pearson correlations among factors are r = 0.21 I–II (physical demands and interpersonal complexity), r = 0.06 I–III (physical demands and substantive complexity), and r = 0.37 II–III (interpersonal complexity and substantive complexity). The proportions of item variance explained by the factors are 20.9% I, 18.0% II, and 15.0% III. 46 dimensions had relatively pure loadings (≥|0.50| on 1 factor and ≤|0.30| on the other 2 factors) on factor I, another 46 on factor II, and an additional 36 on factor III, with 7 additional dimensions having loadings ≥|0.50| on factors I and II, 4 on factors I and III, and 3 on factors II and II. Only 12 dimensions had loadings ≤ |0.40| on all 3 factors.

Principal Components Analysis Within the Conceptual O*NET Subdomains

Another approach to data reduction in civilian studies of the O*NET dimensions has been to work with the conceptual dimensions specified by the developers of the O*NET system, to determine the number of dimensions that can be discovered in this scheme in a first-order PCA/EFA, and then to carry out a second-order PCA/EFA of those first-order scale scores. ²⁵ The purpose of doing this is to see if we can find a more nuanced structure among the 246 O*NET dimensions in this way than by working directly with the 246 primary dimensions. We began this phase of analysis by carrying out separate PCAs for the dimensions within each of the nine subdomains specified by the developers of the O*NET system in the worker characteristics, worker requirements, and occupational requirements domains. These within-subdomain PCAs yielded evidence of between one and three reliable components within each of the nine O*NET subdomains (18 components across all subdomains). (Table II) A brief overview of these components follows, with dimensions having highest standardized partial regression coefficients reported in parentheses.

TABLE II.

Summary of Significant Factors Found in Within-Subdomain Exploratory Factor Analyses of the 246 O*NET Variables in the Army STARRS 2004–2009 Administrative Records Person-Month Sample (n = 100,877)^a

	Number of Dimensions in the Subdomain	Number of Significant Factors	% of Within-Domain Variance Explained	Between-Factor Correlations
				I	II	III
I. Worker Characteristics
A. Abilities	52	2	72
Physical Abilities	33		47	—
Cognitive Abilities	19		25	0.20	—
B. Occupational Interests	6	2	62
Teaching/Leading	3		33	—
Creativity/Intellectual Interests	3		29	0.04	—
C. Work Values	6	1	69
D. Work Styles	16	1	65
II. Worker Requirements
A. Basic Skills	10	1	70
B. Cross-Functional Skills	25	2	70
Organizational Skills	16		39	—
Physical System Skills	9		31	0.01	—
C. Knowledge	33	3	67
Social Sciences	18		44	—
Engineering and Technology	9		15	0.28	—
Business/Management	6		8	0.34	0.22	—
III. Occupational Requirements
A. Generalized Work Activities	41	3	73
Supervision	21		51	—
Interacting With Others	11		14	0.51	—
Technical Complexity	9		8	0.15	0.21	—
B. Work Context	57	3	67
Hazardous Work Conditions	29		43	—
Dealing With Others	19		18	0.45	—
Interpersonal Intensity	9		6	−0.13	0.18	—
Totals	246	18

^aBased on 9 separate exploratory principal axis factor analysis, 1 for all the dimensions in each of the 9 O*NET subdomains. The proportions of item variance in each domain explained by the factors are reported in the third columns. The number of factors to extract in each solution was determined by simulation. Pearson correlations among factors and are reported in the last 2 columns.

Worker Characteristics Subdomains

Parallel analysis suggested that there are four subdomains in the worker characteristics domain. We refer to these as the “abilities,” “occupational interests,” “values,” and “work styles” subdomains based on inspection of standardized partial regression coefficients. Two components were identified for the 52 dimensions in the abilities subdomain: “physical abilities” (e.g., multilimb coordination and speed of limb movements) and “cognitive abilities” (e.g., oral expression and deductive reasoning). Two components were also identified for the 6 dimensions in the occupational interest subdomain: “teaching/leading” (e.g., social and enterprising) and “creativity/intellectual interests” (e.g., investigative and artistic). Parallel analysis of the 6 dimensions in the work values subdomain (e.g., achievement and recognition) and the 16 dimensions in the work styles subdomain (e.g., initiative and persistence) suggested that these subdomains are one-dimensional.

Worker Requirements Subdomains

Parallel analysis suggested that the worker requirements domain has three subdomains: “basic skills,” “cross-functional skills,” and “knowledge.” The 10 dimensions in the basic skills subdomain appear one-dimensional (e.g., critical thinking and learning strategies). Two components were identified for the 25 dimensions in the cross-functional skills subdomain: “organization skills” (e.g., instructing and judgment/decision making); and “physical system skills” (e.g., troubleshooting and equipment selection). Three components were identified for the 33 dimensions in the knowledge subdomain: “social science” (e.g., therapy/counseling and psychology), “engineering and technology” (e.g., engineering/technology and physics), and “business/management” (e.g., administration management and economics-accounting).

Occupational Requirements Subdomains

Parallel analysis suggested that the occupational requirements domain has two subdomains: “generalized work activities” and “work context.” Three components were identified for the 41 dimensions in the generalized work activities subdomain: “supervision” (e.g., monitoring and controlling resources; guiding, directing, and motivating subordinates); “interacting with others” (e.g., performing for or working directly with the public and identifying objects, actions, and events); and “technical complexity” (e.g., controlling machines and processes; repairing and maintaining mechanical equipment). An additional three components were identified for the 57 dimensions in the work context subdomain: “hazardous work conditions” (e.g., being exposed to high places and to uncontrolled environmental conditions); “dealing with others” (e.g., dealing with external customers and working with group teams); and “interpersonal intensity” (e.g., electronic mail and telephone).

Second-Order Principal Components Analysis of Subdomain Component Scores

Component-weighted scores were constructed for the 18 scales in the within-subdomain PCAs, and a second-order PCA of those component scores was carried out to examine whether we could recover the three components derived in the aggregate PCA. Parallel analysis suggested that there are, in fact, three components underlying the 18 component-weighted scales in the subdomain PCAs. These three correspond closely to the three component scores found in the aggregate PCA of all 246 individual O*NET dimensions (Table III). Pearson correlations between scores on the three O*NET components based on the aggregate PCA of the 246 dimensions and comparable scores on the second-order PCA of the within-subdomain first-order component scales are r = 0.95 for physical demands, r = 0.92 for interpersonal complexity, and r = 0.90 for substantive complexity. Because specification of all 21 of these variables as predictors in a single equation would yield highly unstable coefficients as a result of the strong three-component structure underlying the 18 domain-specific scales, we regressed scores on each of the subdomain 18 scales onto the three domain scores to generate 18 residual subdomain component scores that would be independent of the three main components.

TABLE III.

Standardized Partial Regression Coefficients of the 18 Factor-Weighted Scales Obtained From Within-Domain Exploratory Factor Analyses of the O*NET Dimensions Based on Three-Factor Promax Rotated Second-Order Exploratory Factor Analysis of These Scores in the Army STARRS 2004–2009 Administrative Records Person-Month Sample (n = 100,877)^a

	Physical Demands	Interpersonal Complexity	Substantive Complexity
I. Worker Characteristics
A. Abilities
Physical Abilities	0.63	0.68	−0.29
Cognitive Abilities	0.04	0.31	0.77
B. Occupational Interests
Teaching/Leading	−0.61	0.49	0.38
Creativity/Intellectual Interests	0.49	0.14	0.38
C. Work Values	0.09	0.55	0.54
D. Work Styles	0.04	0.78	0.21
II. Worker Requirements
A. Basic Skills	−0.03	0.28	0.82
B. Cross-Functional Skills
Organizational Skills	0.05	0.18	0.85
Physical System Skills	0.94	−0.26	0.10
C. Knowledge
Social Sciences	−0.06	0.82	0.20
Engineering and Technology	0.88	−0.21	0.47
Business/Management	0.06	−0.06	0.75
III. Occupational Requirements
A. Generalized Work Activities
Supervision	0.21	0.01	0.90
Interacting With Others	−0.01	0.88	0.10
Technical Complexity	0.94	0.05	−0.10
B. Work Context
Hazardous Work Conditions	0.81	0.39	−0.17
Dealing With Others	−0.04	0.97	−0.09
Interpersonal Intensity	−0.22	−0.16	0.83

^aBased on second-order exploratory principal axis factor analysis with promax rotation of 18 first-order factor-weighted O*NET subdomain factor scores. The number of factors to extract was determined by simulation. All boldfaced coefficients are significant p < .001. Pearson correlations among factors are r = 0.15 physical demands and interpersonal complexity, r = 0.03 physical demands and substantive complexity, and r = 0.43 interpersonal complexity and substantive complexity. The proportions of item variance explained by the factors are 11% physical demands, 24% interpersonal complexity, and 46% substantive complexity.

Associations of Rank With the 21 O*NET Dimensions

As assignment of MOS/AOCs is contingent on a soldier’s rank, we were interested in examining the associations of rank/grade with the 3 domain and 18 subdomain scores. To do this, one-way analysis of variance was used to examine associations of a seven-category rank/grade classification scheme (E1–E2; E3–E4; E5–E9; Warrant Officer; Commissioned Officer Lieutenant I–II; Commissioned Officer Captain Major; and Commissioned Officer Lieutenant Colonel and above) with each of the three domain scores and 18 residual subdomain scores. Results showed that rank/grade account for only small proportions of variance in the component scores, with a mean ε² of 0.03, an ε² range of 0.001–0.175, and with 20 of the 21 models having an ε² < 0.10. These results suggest that the majority of the information in the O*NET component scores reflects variation within rather than between these broadly defined rank-grade groupings (Table IV).

TABLE IV.

Means and Standard Deviations Obtained from Multiple Linear Regressions of 21 O*NET Component Scores^a onto 7 Categories of Rank in the Army STARRS 2004–2009 Administrative Records Person-Month Sample (n = 100,877)

Component (Outcomes)	Rank (Predictors)
	E1–E2		E3–E4		E5–E9		Warrant Officer		Commissioned Officer Lt. I–IIb		Commissioned Officer Captain/Major		Commissioned Officer Lt. Col. or Abovec		Model ε2
	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD
I. Worker Characteristics
A. Abilities
Physical Abilities	−0.02	0.24	−0.01	0.24	−0.01	0.25	0.19	0.26	0.09	0.25	0.07	0.26	0.03	0.26	0.024
Cognitive Abilities	0.02	0.24	0.03	0.25	0.00	0.27	0.13	0.42	−0.14	0.36	−0.12	0.35	−0.16	0.36	0.034
B. Occupational Interests
Teaching/Leading	−0.06	0.43	−0.04	0.42	0.02	0.46	0.13	0.47	0.18	0.54	0.07	0.63	0.09	0.62	0.013
Creativity/Intellectual Interests	0.09	0.56	0.07	0.60	−0.06	0.70	−0.17	1.04	−0.33	0.72	0.03	1.08	0.11	1.29	0.014
C. Work Values	−0.03	0.33	−0.02	0.34	−0.04	0.33	0.46	0.76	0.11	0.33	0.17	0.36	0.17	0.36	0.066
D. Work Styles	−0.03	0.38	−0.03	0.39	0.01	0.42	0.59	0.53	−0.12	0.37	−0.03	0.47	−0.01	0.49	0.051
II. Worker Requirements
A. Basic Skills	−0.01	0.21	0.00	0.21	−0.01	0.21	0.16	0.28	0.01	0.26	0.03	0.31	0.00	0.35	0.014
B. Cross-Functional Skills
Organizational Skills	−0.05	0.31	−0.05	0.31	0.00	0.35	0.12	0.33	0.20	0.28	0.17	0.34	0.14	0.40	0.038
Physical System Skills	0.02	0.43	0.03	0.43	0.02	0.45	−0.11	0.61	−0.29	0.32	−0.17	0.34	−0.10	0.33	0.024
C. Knowledge
Social Sciences	0.03	0.26	0.02	0.27	−0.03	0.28	0.03	0.52	0.03	0.22	0.00	0.35	0.01	0.46	0.006
Engineering and Technology	0.08	0.41	0.05	0.39	−0.01	0.41	0.17	0.41	−0.27	0.66	−0.27	0.64	−0.20	0.66	0.046
Business/Management	0.01	0.68	0.02	0.69	0.06	0.75	−0.24	1.05	−0.20	0.79	−0.31	0.96	−0.20	0.96	0.019
III. Occupational Requirements
A. Generalized Work Activities
Supervision	0.01	0.27	0.00	0.27	0.02	0.29	−0.30	0.43	0.00	0.36	−0.06	0.42	−0.03	0.42	0.028
Interacting With Others	0.02	0.33	0.02	0.34	0.01	0.37	−0.27	0.39	−0.10	0.22	−0.06	0.27	−0.06	0.30	0.021
Technical Complexity	0.00	0.21	0.01	0.22	0.00	0.22	−0.02	0.16	−0.03	0.21	0.01	0.26	−0.02	0.30	0.001
B. Work Context
Hazardous Work Conditions	0.00	0.21	−0.01	0.22	0.00	0.22	−0.05	0.18	0.06	0.26	0.03	0.30	0.06	0.32	0.005
Dealing With Others	0.02	0.29	0.02	0.31	0.00	0.30	−0.31	0.44	0.01	0.23	0.01	0.34	−0.04	0.40	0.025
Interpersonal Intensity	0.01	0.67	0.00	0.68	0.03	0.65	0.23	0.36	−0.15	0.60	−0.23	0.67	−0.15	0.69	0.013
IV. Higher-Order Components
A. Interpersonal Complexity	−0.04	1.03	−0.09	1.02	−0.04	0.99	0.06	0.50	0.37	0.68	0.40	0.62	0.32	0.59	0.020
B. Physical Demands	0.27	0.88	0.18	0.93	−0.02	1.01	−0.23	0.76	−0.46	0.97	−0.85	0.96	−1.14	0.90	0.093
C. Substantive Complexity	−0.15	0.96	−0.21	0.94	−0.12	1.00	0.57	0.61	1.16	0.49	1.19	0.45	1.23	0.44	0.175
n	8,984		36,901		42,261		2,368		2,190		6,182		1,991

^aAll component scores were initially standardized to a mean of 0 and a variance. The 18 subdomain scores were then regressed on the 3 component scores (Interpersonal complexity, Physical demands, Substantive complexity) and the residuals of these subdomain scores were used as outcomes in this table. This means that variances of the residual subdomain scores vary but are consistently less than 1.0, while variance of the domain scores is 1.0.

^bLt. I–II = first or second lieutenant.

^cLt. Col. = lieutenant colonel.

DISCUSSION

Our aim in this study was to identify meaningful summary measures of military occupational characteristics that could be used in future research to examine associations of occupational characteristics with diverse soldier outcomes (e.g., job satisfaction, work performance, health). The results are broadly consistent with the two civilian studies in which PCA/EFA was estimated with the full set of O*NET dimensions.^25,26 Both civilian studies, consistent with our study, found three components/factors underlying the O*NET dimensions. In addition, one of those two earlier studies carried out a second-order EFA and found the same three-factor structure as we did among the first-order O*NET subdomain factor scores.²⁵ Furthermore, a number of previous PCA/EFA studies of individual O*NET subdomains²⁵ and more limited sets of O*NET dimensions^{20,24,33–35} found within-subdomain components/factors were similar to most of those identified here.

Despite these broad consistencies, three noteworthy differences exist between our results and the civilian results. First, although psychomotor abilities were the most prominent items in civilian studies of the physical demands factor, this factor is dominated by hazardous work conditions in Army MOS/AOCs. This difference is likely as a result of the fact that the two civilian EFA studies that analyzed the full set of O*NET dimensions both treated the O*NET occupational groupings as the unit of analysis, thus giving equal weight to each grouping no matter how common it was in the civilian labor force, whereas our analysis, as noted in the section on sampling, mapped the O*NET dimensions onto the population of all active duty soldiers, implicitly weighting the data by the distribution of MOS/AOCs in the Army. The greater prominence of hazardous work conditions in our analysis, then, reflects the greater importance of these conditions among Army occupations than civilian occupations.

Second, the interpersonal complexity component in our analysis is different from the comparable factor in the civilian studies in that our interpersonal complexity component consists more exclusively of O*NET dimensions related to interpersonal aspects of work and includes more complex characteristics, such as social challenges and conflict situations, than the civilian factor; whereas, the civilian factor not only has strong positive loadings on O*NET interpersonal complexity dimensions but also strong negative loadings on O*NET dimensions involving use of technical equipment. In the Army sample, the latter dimensions loaded more strongly on the physical demands component. We suspect that these differences are due to differences in the nature of Army occupations, with the interpersonal complexity of Army occupations not having the same inverse relationships to the technical complexity of work as in the civilian labor force.

Third, consistent with the last point, the intercomponent correlations involving the physical demands component were different in two respects in our study compared to the one civilian study where these correlations were reported.²⁵ The first is that although the physical demands and interpersonal complexity components inversely correlated in the civilian labor force (all variables factor scores ε = −0.25; subdomain factor scores ε = −0.14),²⁵ they are positively correlated among Army MOS/AOCs (all variables component scores ε = 0.21; subdomain component scores ε = 0.18). This means that civilian occupations with high interpersonal complexity are less likely than other occupations to be physically demanding, whereas Army MOS/AOCs with high interpersonal complexity are more likely than others to be physically demanding. This difference reflects the fact that Army leaders are more likely than civilian leaders to “lead from the front.” Similarly, the correlation of physical demands with substantive complexity is more negative in the civilian data (all variables factor scores r = −0.16; second-order subdomain factor scores ε = −0.29)²⁵ than in the Army (all variables component scores ε = 0.06; subdomain components scores ε = −0.03), reflecting the fact that civilian jobs requiring complex problem solving and creative thinking are less likely than Army MOS/AOCs requiring comparable skills to involve exposure to hazardous work conditions.

As noted in the introduction, prior literature on differences in health outcomes across military occupations has relied for the most part on coarse categorical schemes for classifying MOS/AOCs based on broad similarities in occupational functions (e.g., distinguishing combat arms from other occupations).^11,14,36 Although such schemes might be sufficient for studying some soldier outcomes, the O*NET scales derived here make it possible to augment these broad occupational classification schemes to investigate more subtle characteristics of differences in job conditions across MOS/AOCs related to a wide range of soldier outcomes. We have posted the 21 summary scales derived here at (http://www.armystarrs.org/publications) for use by other researchers.

These 21 scales could be used in a number of ways. For one, they could be used as the basis of a cluster analysis³⁷ of commonly occurring multivariate job profiles that might be treated as predictors of adverse outcomes that have been previously studied in both military and civilian occupations, such as substance use, obesity, hypertension, physical injury, hospitalization, disability discharge, and suicide.^{11,12,14,20,36} Another possibility would be to use the 21 scales in nested discretized form as the starting point for regression tree analyses aimed at detecting multivariate job condition profiles associated with high risk of adverse outcomes.³⁸ This is a particularly attractive option for studying effects of job conditions on health given that much of the theorizing about these effects, most notably the influential demand–control model⁵ and its extensions,³⁹ hypothesize that mismatch between multiple job conditions is centrally involved in the adverse health effects of pathogenic occupations.

Several limitations of this study should be noted. First, errors may have occurred in selecting composite civilian occupational groups to characterize MOS/AOCs that lack civilian equivalents. Second, the decision to use maximum O*NET dimension scores across these composite occupational groups rather than alternatives (e.g., means; unique ratings assigned by Army occupational experts) might have affected results. Third, additional imprecision could have been introduced by O*NET scores being imperfect, as this scoring is based on rater judgments and therefore less fine-grained than worker reports of actual job conditions,⁴⁰ leading to conservative bias in estimates of associations between actual job conditions and outcomes.²⁶ Fourth, even if O*NET ratings are accurate for prototypic civilian job contexts, it might be that the job conditions for some occupations are different in the Army than the civilian labor force. Fifth, given that our analysis was based on a person-month dataset rather than a dataset of 610 MOS/AOCs, results were weighted to think of individuals as the unit of analysis rather than to think of occupations as the unit of analysis. We considered this a preferred approach in that our interest is in using these scales to understand individual differences (rather than occupational differences) in health outcomes, but it needs to be noted that the results might be suboptimal for carrying out research in which occupations rather than individuals are of primary interest.

Despite these limitations, this study usefully extends O*NET classifications to Army MOS/AOCs. Future research is needed to determine whether the scales derived here have similar associations with mental and physical health outcomes as those found in civilian studies.^41–43 If so, it would be useful to extend the analyses to evaluate the extent to which other aspects of Army service (e.g., length of deployment, time in service, rank) may moderate these relationships. Such research could be valuable in helping target health promotion and risk reduction efforts to soldiers.