Table 4.
Effects of weight trimming on the bias‐efficiency trade‐off for selected outcome variable prevalence estimates in the three main Army STARRS survey samplesa
| New Soldier Study Q2–4 2011b | All Army Study Q2–4 2011b | Baseline Pre‐Post Deployment Study | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Trimming pointc | Trimming pointc | Trimming pointc | |||||||
| 2 | 5 | 10 | 2 | 5 | 10 | 2 | 5 | 10 | |
| Generalized anxiety disorder | 95.9 | 85.3 | 72.7 | 98.7 | 121.7 | 192.5 | 100.2 | 100.6 | 98.5 |
| Intermittent explosive disorder | 99.8 | 102.6 | 82.0 | 109.1 | 99.4 | 101.8 | 100.1 | 100.7 | 100.2 |
| Major depressive episode | 105.1 | 90.9 | 87.8 | 116.9 | 65.5 | 76.6 | 98.3 | 97.5 | 95.4 |
| Panic disorder | 101.4 | 93.4 | 94.4 | 102.9 | 93.5 | 101.9 | 99.2 | 99.7 | 97.0 |
| Post‐traumatic stress disorder | 100.6 | 111.1 | 120.2 | 99.7 | 138.2 | 172.4 | 97.2 | 95.9 | 94.9 |
| Suicide ideation | 101.2 | 100.6 | 148.6 | 96.1 | 100.2 | 148.2 | 95.6 | 95.3 | 92.3 |
| Any 30‐day disorder | 99.8 | 89.1 | 94.2 | 102.9 | 102.3 | 96.0 | 99.7 | 99.6 | 97.4 |
| (n) | (11,802) | (5428) | (7425) | ||||||
The samples were doubly weighted to adjust for differences on SAQ variables between SAQ respondents who consented and provided linking information for ADS data versus those that did not (WT1) and between the weighted (WT1) sample of SAQ respondents with linked ADS data and the population (WT2).
The NSS and AAS studies were piloted in Q1 2011 absent the questions about suicidality and the safety plan associated with those questions (which did not receive IRB approval until after the Q1 replicates were fielded). Full implementation started in Q2 2011, which is why this was the first replicate included in the weighting. Data for 2012 are not reported here because weighting of the 2012 NSS and AAS data are being carried out separately using an updated population post‐stratification dataset for that year.
The trimming point is the proportion of respondents trimmed at each tail of the distribution. Four per cent of respondents (2% at the upper end of the distribution and 2% at the lower end of the distribution) were trimmed in the solution with a trimming point of 2, 10% at a trimming point of 5, and 20% at a trimming point of 10. See the text for a more detailed description of the procedures and rationale for trimming. Results for trimming points at each whole number value in the range 1–10 are available on request.