Re: WEIGHT SAMPLING REPRESENTATIVITY [message #8365 is a reply to message #8364] |
Thu, 15 October 2015 19:13 ![Go to previous message Go to previous message](/theme/default/images/up.png) |
Reduced-For(u)m
Messages: 292 Registered: March 2013
|
Senior Member |
|
|
100 people who have chosen to consume a product can never represent the entire population (probably not even the entire population that consumes the product). That may or may not be a huge problem in your analysis - you don't need that sub-group to be population-representitive. But any analysis on whether or not use of some product affects labor supply would need to include the observations of those that do NOT consume it anyway, so as to have a comparison rate of labor force participation. If you want to do this analysis knowing there are only 100, self-selected people who use the product, you can (though you are not likely to have sufficient precision to get a statistically significant result), but you should include everyone in the sample in your analysis, and compare those that did take it and those that did not.
When you use the whole sample, you should use the regular weights. Also, which types of weights are you specifying them as? If you are using them as "frequency weights", which I'm guessing might be the case, that would be the wrong type. The DHS weights are modified probability weights: in Stata for example you could specify them as "aweight" (aalytics weights) or "pweight"(probability weights).
|
|
|