Know-How Exchange : Research/Metrics

Your proposed calculation sounds right to me. Just be sure that every respondent is weighted so the total of weighted responses is 371.

One of the risks is that when you sub-segment the respondent base as much as you have you have some very small cells, so that the reliability of the weighted number is suspect. As long as you recognize that, you should be OK.

Generally speaking, the conclusions don't change much when you use weighted data vs. unweighted data, so look at results both ways.

What you're doing sounds correct. Say you have 40:60 male to female ratio and you want the final total to be 50:50, then you multiple the males by 50/40 = 0.8 and multiple the females by 50/60 = 1.2 to get the final answer.

If you are looking at results which are a combination of different groups its best to apply the weight to each person in your sample individually then aggregate by adding up the weighted values and dividing by the weighted total as this makes the analysis far easier.

So using the example above, if you're looking at 'Owns a car', instead of counting '1' as 'Yes' for each man, you count with a value 0.8, and instead of counting '1' for each woman you count using 1.2.

Say you have 15 men out of 20 saying they own a car, and 12 women out of 24. You want to know the weighted percentage of ownership.

Total up 15 lots of 0.8 for men plus twelve lots of 1.2 for women - ie 12+14.4=26.4 as weighted car owners. This is then out of a weighted total possible of 20 lots f 0.8 for men and 24 lots of 1.2 for women - ie 16+28.8 = 44.8. So the weighted percentage is 26.4/44.8 = 58.9%

I think there's something wrong with Saul's calculation. He has flipped the ratio of men to women in his example.

If 3/4 of the men own cars and 1/2 the women own cars, and if you want to know what percentage of a 50/50 population own cars (based on your sample), it should be 62.5%. (3/4 of 50% men, plus 1/2 of 50% women = 62.5% of total population)