Skip to main content

Table 2 Model fit statistics and model comparison tests (n = 18,838 men and 2952 women)a,b

From: Improving risk prediction accuracy for new soldiers in the U.S. Army by adding self-report survey data to administrative data

  

Male physical violence perpetration

Male sexual violence perpetration

Female sexual violence victimization

df

χ2

p

df

χ2

p

df

χ2

p

I. Modelsc

 M1

T

32

93.4

<.0001

34

47.7

0.059

29

498.1

<.0001

 M2

T + A

33

638.4

<.0001

35

152.8

<.0001

30

521.7

<.0001

 M3

T + A + T*A

35

711.0

<.0001

37

185.5

<.0001

32

851.3

<.0001

 M4

T + A + A2

34

497.2

<.0001

36

99.5

<.0001

31

549.7

<.0001

 M5

T + A + T*A + A2

36

524.9

<.0001

38

138.4

<.0001

33

866.1

<.0001

 M6

Best model for A (Ba) + S

34

578.5

<.0001

36

268.9

<.0001

31

473.4

<.0001

 M7

Ba + S + T*S

36

597.2

<.0001

38

308.8

<.0001

33

574.7

<.0001

 M8

Ba + S + S2

35

576.5

<.0001

37

253.3

<.0001

32

635.5

<.0001

 M9

Ba + S + T*S + S2

37

604.7

<.0001

39

268.9

<.0001

34

678.9

<.0001

 M10

Ba + S + A*S

–

–

–

37

238.1

<.0001

–

–

–

 M11

Ba + S + A*S + T*S

–

–

–

39

291.5

<.0001

–

–

–

II. Model Differences

 M2-M1

A

1

259.2

<.0001

1

42.8

0.000

1

15.3

0.000

 M3-M2

T*A

2

1.5

0.469

2

2.2

0.339

2

0.6

0.744

 M5-M4

T*A

1

3.0

0.085

1

2.9

0.089

1

0.2

0.690

 M4-M2

A2

2

2.1

0.351

2

2.3

0.324

2

0.3

0.848

 M5-M3

A2

1

3.9

0.050

1

3.2

0.073

1

0.0

0.959

 M6-Ba

S

1

24.2

0.000

1

54.1

0.000

1

43.3

<.0001

 M7-M6

T*S

2

0.5

0.797

2

6.8

0.034

2

0.3

0.871

 M9-M8

T*S

1

0.4

0.543

1

0.4

0.530

1

0.2

0.629

 M8-M6

S2

2

0.5

0.796

2

6.0

0.050

2

0.3

0.877

 M9-M7

S2

1

0.4

0.527

1

1.3

0.253

1

0.3

0.616

 M11-M10

T*S

–

–

–

2

5.9

0.053

–

–

–

 M11-M7

A*S

–

–

–

1

3.6

0.059

–

–

–

  1. Abbreviations: Time (T) time since survey administration (main effects of T dummy coded with each month), S predicted log odds from New Soldier Survey (NSS), A predicted log odds from Historical Administrative Data System (HADS), A2 the square of A, T*A the interaction between T and A (where T is dummy coded with indicator variables for 13–24 months and 25+ months), Ba predictors from best model among models 1 through 5, T*S interaction between T and S (T dummy coded with indicator variables for 13–24 months and 25+ months), S2 S-squared, S*A interaction of S and A
  2. aThe NSS respondents considered here were surveyed between April 2011 and November 2012. Administrative data were available through December 2014 (25-44 months after the survey). The sample size decreased with duration both because of attrition and because of variation in time between survey and end of the follow-up period. The sample included 18,838 men (decreasing to 16,479 by 12 months, 15,306 by 24 months, and 3,729 by 36 months) and 2,952 women (decreasing to 2,300 by 12 months, 2,094 by 24 months, and 687 by 36 months).
  3. bAlthough the same sample of soldiers was used for both male outcomes, the number of person-months differed because we predicted first occurrence of each outcome, and each soldier was censored after the month when the outcome first occurred, termination of service, or December 2014, whichever came first. Number of person-months was 543,603 for male physical assault perpetration, 543,636 for male sexual assault perpetration, and 75,772 for female sexual assault victimization.
  4. cOut of M1-M5, M2 was the best model for each outcome; M6-M11 add NSS predicted log odds to the best model (Ba) from HADS data alone. The final best models were M6 for physical violence perpetration and sexual violence victimization and M7 for sexual violence perpetration