Online supplement Table S1 evaluates the impact that

advertisement
Online supplement
Table S1 evaluates the impact that misclassification would have on the estimated relative
risk of violence among inmates with histories of trauma. The table is broken into four sections.
The first row, shows the actual data (i.e. assuming that all incidents were reported). The 2nd
portion of the table shows scenarios where under-reporting is equally likely for those with and
without trauma histories. For example, in the first scenario, only 25% of all incidents are
detected for inmates with or without histories of trauma. As seen in the fourth column, the
relative risk is the same, despite the fact that the actual incidence of violence is 4 times higher
than what was reported. This does not change, regardless of the amount of under-reporting as
shown in the 3 scenarios that follow.
The third portion of the table shows scenarios where under-reporting is more common
among inmates with histories of trauma. In all cases, the true relative risk (if all incidents were
captured) is higher than what is estimated in this study (i.e. always greater than 1.8). The greater
the difference in reporting rates between those with and without trauma, the greater the true
relative risk would be. Conversely, the fourth portion of the table shows scenarios where underreporting is more common among those without a trauma history. In this case, the true relative
risk is lower than estimated in the study (i.e. is less than 1.8). The greater the difference in underreporting between those with and without trauma, the lower the true relative risk would be. In
some extreme cases, where the under-reporting rate for those without trauma is more than 25%
lower than it is among those with trauma, the risk of violence would actually be less among those
with trauma histories.
Table S1. Assessment of potential bias introduced due to under-reporting of incidents.
% incidents reported
True proportion of inmates with incidents
Trauma
No trauma
Trauma
No trauma
RR
Actual data
100%
100%
7.4%
4.1%
1.80
Under-reporting is equally likely regardless of whether the risk factor is present
25%
25%
29.6%
16.4%
1.80
40%
40%
18.5%
10.3%
1.80
60%
60%
12.3%
6.8%
1.80
75%
75%
9.9%
5.5%
1.80
25%
25%
25%
25%
40%
40%
40%
40%
50%
50%
50%
60%
60%
Under-reporting more likely if the risk factor is present
30%
29.6%
13.7%
40%
29.6%
10.3%
50%
29.6%
8.2%
75%
29.6%
5.5%
45%
18.5%
9.1%
50%
18.5%
8.2%
60%
18.5%
6.8%
75%
18.5%
5.5%
55%
14.8%
7.5%
60%
14.8%
6.8%
75%
14.8%
5.5%
70%
12.3%
5.9%
75%
12.3%
5.5%
2.17
2.53
3.61
5.41
2.03
2.26
2.71
3.38
1.99
2.17
2.71
2.11
2.26
40%
50%
50%
60%
60%
60%
75%
75%
75%
75%
Under-reporting more likely if the risk factor is absent
25%
18.5%
16.4%
25%
14.8%
16.4%
40%
14.8%
10.3%
25%
12.3%
16.4%
40%
12.3%
10.3%
50%
12.3%
8.2%
25%
9.9%
16.4%
40%
9.9%
10.3%
50%
9.9%
8.2%
60%
9.9%
6.8%
1.13
0.90
1.44
0.75
1.20
1.50
0.60
0.96
1.20
1.44
Download