The trouble with you example of death penalty is that the sample size is 10x difference between white and black. Perhaps using weighted average would be better?
That could work indeed for the overall table. Basically, as you suggest, per type of victim (B or W) one could calculate weights so that W defendants and B defendants are counted stronger or weaker so that it corrects for the fact that in the example W defendants were more likely to have killed W victims. Looking at W victims (min 8:25), there are 151 W defendants, but only 63 B defendants. One could give the B defendants a weight of 151/63. This weighs W and B defendants equally. So instead of 11 and 52, we would use 11 * 151/63 and 52 * 151/63. This does not change the percentages at min 8:25, but it would change the overall table. Similarly, one would have to do this for B victims (min 8:49). Here one way to weigh W and B defendants equally would be to weigh W defendants with 106/9. So instead of 0 and 9, we would use the numbers 0 * 106/9 and 9 * 106/9. Again, this does not change the percentages here, but it would for the overall table when these numbers are used that adjust for W or B defendant. Of course, then you could even adjust for the fact that there were more W victims than B victims. Now the weight vor W victims would be (9 + 6 + 67)/ (19 + 132 + 11 + 52). You would then combine these weights for calculating the overall table.
Amazing ! Thank you .
Glad the video is helpful
the second example is not eco fallacy. It is due to confounder. No leveling here .
The trouble with you example of death penalty is that the sample size is 10x difference between white and black. Perhaps using weighted average would be better?
That could work indeed for the overall table. Basically, as you suggest, per type of victim (B or W) one could calculate weights so that W defendants and B defendants are counted stronger or weaker so that it corrects for the fact that in the example W defendants were more likely to have killed W victims. Looking at W victims (min 8:25), there are 151 W defendants, but only 63 B defendants. One could give the B defendants a weight of 151/63. This weighs W and B defendants equally. So instead of 11 and 52, we would use 11 * 151/63 and 52 * 151/63. This does not change the percentages at min 8:25, but it would change the overall table. Similarly, one would have to do this for B victims (min 8:49). Here one way to weigh W and B defendants equally would be to weigh W defendants with 106/9. So instead of 0 and 9, we would use the numbers 0 * 106/9 and 9 * 106/9. Again, this does not change the percentages here, but it would for the overall table when these numbers are used that adjust for W or B defendant. Of course, then you could even adjust for the fact that there were more W victims than B victims. Now the weight vor W victims would be (9 + 6 + 67)/ (19 + 132 + 11 + 52). You would then combine these weights for calculating the overall table.