The simplest way to deal with a small significant group (having a significant count of fatalities, such as alcoholics) within the car driver population is to treat the very small group as a stream of negligible size and factor that stream out completely.
This page demonstrates that significant small streams can be discarded without significantly affecting the result.
A PRIORI
|
|
|
|
|
2/(1+2) |
|
|
|
|
|
|
|
|
Columns 1, 2, 3 & 4 contain fatality counts. Columns 5, 6 & 7 were calculated from columns 1 . . 4 according to the formula given in the column header.
Calculation of the wearing rates and effectiveness using two streams:
Stream A of assumed size 1% of the population. It can be deduced
that alcoholics would be wearing belts 25.3% of the time to produce the
observed fatality rate of 30.5%.
Stream B will consist of the remaining 99% of the population.
It can be deduced that people who are not alcohol affected were wearing
belts 65.5% to produce the observed rate at fatality of 71.1%.
When these two streams are combined, their combined (average) belt wearing rate would be
0.01 * 25.3% + 0.99 * 65.5% = 65.1% (as found by NOPUS)
Effectiveness is given by the expression E = 1-(b/u)*(1-p)/p {where b=belted, u=unbelted, p=proportion wearing belts}
So the belt effectiveness for stream A is
E = 1 - 1,376/3,139*(1-0.253)/0.253
= -29.4%
And the effectiveness for stream B is
E = 1 - 10,254/4,174*(1-0.655)/0.655
= -29.4%
Calculation of the wearing rates and effectiveness using the major stream:
Wearing rate of stream B = 65.1%
Effectiveness of stream B is
E = 1 - 10,254/4,174*(1-0.651)/0.651
= -31.7%
SUMMARY
With increases in the assumed size of the alcohol stream, the calculated
value of seat belt Effectiveness becomes more positive.
If the proportion of alcohol affected drivers was assumed to be about
15% of all drivers, the seat belt effectiveness would be zero.
Ignoring the small stream if it was less than 1% does not significantly effect the result.
Tabulation of calculations with group size as the variable is done on the data & calculations document in Tables 6a and 6b.