Skip to main content

Table 4 Performance of univariate and best-fitting models (n = 18,838 men and 2952 women)a

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

 

Top ventile (5%)

Top two ventiles (10%)

Top three ventiles (15%)

HADS-only

NSS-only

Bestb

Proportional Improvement Best/HADSc

HADS-only

NSS-only

Bestb

Proportional Improvement Best/HADSc

HADS-only

NSS-only

Bestb

Proportional Improvement Best/HADSc

I. Concentration of Risk (%)

 Male physical violence perpetration

33.9

24.8

39.5

16.6

45.2

38.4

50.2

11.2

52.3

48.7

56.3

7.8

 Male sexual violence perpetration

20.7

21.8

26.1

26.0

32.4

33.8

42.0

29.6

35.5

46.5

51.8

45.9

 Female sexual violence victimization

17.5

27.6

29.4

67.9

32.1

38.3

41.3

28.7

47.6

47.9

49.8

4.8

II. Observed Positive Predictive Valued

 Male physical violence perpetration

2.9

2.2

3.4

17.2

2.0

1.7

2.2

10.0

1.5

1.4

1.6

6.7

 Male sexual violence perpetration

1.2

1.2

1.5

25.0

0.9

1.0

1.2

33.3

0.7

0.9

1.0

42.9

 Female sexual violence victimization

6.8

10.6

11.5

69.1

6.3

7.4

8.1

28.6

6.2

6.2

6.5

4.8

III. Projected Positive Predictive Valuee

 Male physical violence perpetration

68.2

50.4

79.1

16.0

46.0

39.2

51.0

10.9

35.5

33.1

38.4

8.2

 Male sexual violence perpetration

27.6

29.0

34.6

25.4

21.6

22.6

27.9

29.2

15.8

20.7

23.0

45.6

 Female sexual violence victimization

151.6

225.9

241.7

59.4

139.8

163.8

176.4

26.2

138.3

139.0

144.6

4.6

  1. Abbreviations: NSS-only prediction from model based on survey data alone, HADS-only prediction from model based on administrative data alone
  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 3729 by 36 months) and 2952 women (decreasing to 2300 by 12 months, 2094 by 24 months, and 687 by 36 months)
  3. bAdditive model for physical violence perpetration and sexual violence victimization; model including interaction with time for sexual violence perpetration
  4. cProportional increase in concentration of risk or positive predictive value of the best model relative to the HADS-only model
  5. dObserved cases per 1000 person-months
  6. eNumber of cases per 1000 soldiers projected to 36 months