Because "police reported alcohol" is subject to more robust data collection
procedures than seat belt status I have extracted and allocated "police
reported alcohol" in the first instance. Below I argue that ignoring
"belt status Unknown" data produces a conservative result.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1995 |
|
|
|
|
|
1995 |
|
|
|
|
|
1996 |
|
|
|
|
|
For about 64% of fatalities the police reported that alcohol was not involved. In about 6% of cases the presence of alcohol was reported. In about 20% of fatalities the presence or absence of alcohol was not reported, and in about 10% of fatalities the presence of alcohol was reported as "unknown". I have previously (See Table 3 Notes 1..4) distributed those last two data elements.
In all cases, about 48% of belted "Unknown %" is in the "No Alcohol" column, and about 6% of belted "Unknown %" is in the "Alcohol" column, leaving about 25% in the "not reported" category, and 20% in the "unknown" category.
In summary, the belted status is most likely to be unknown if alcohol status is unknown or is not reported.
I have previously argued that the alcohol "not reported" data is an unbiassed sample of the parent population, and that the alcohol "unknown" is a biassed sample of the parent population, in that it contains approximately 50% alcohol affected fatalities.
From the above it can be inferred that the belt status "unknown" group contains a proportion of alcohol affected drivers which is larger than that found in the parent population of alcohol affected drivers, so ignoring it would result (after application of Simpson's Paradox correction) in a conservative estimate of the proportion of belted fatalities.