What number does the sum of the factor of expansion add up to? Total population?

Jeanne -

You say that "The factor suppose to show the number of persons that each observation represents..." Well, say you are using equal probability selection from a population or from a given stratum or subpopulation. Say the total number of people in there is N, and from them you randomly select n. Then, with such an equal probability of selection scheme, the probability of selection for each person is n/N. Therefore each person represents N/n people. Say N = 100, and n = 20. Then the probability of selection is 0.2, and each of the 20 members of the sample will represent 5 in the population, meaning themselves and 4 others. But this does not have to be an integer. It likely would not be. Say N = 269, and n = 25. You know that 269 divided by 25 is not an integer. This is fine. I don't know why there would be a restriction in STATA to an integer. I don't know STATA, and do not know what is meant here by f weights and p weights. I am familiar with survey weights based on probabilities of selection, and regression weights, and calibration weights in surveys which may be a combination of the two, and have various restrictions, but I do not know what would restrict anything to integers. I worked with continuous data, but even if you aren't, I do not know what would be the cause of your problem.

Suggestion: Find a STATA manual, there may be some online, and see if you are using the wrong program routine. I would also think that you could contact STATA with questions.

Best wishes - Jim

Hello Jeanne,

If your sample was randomly selected, then using weights is a great tool for generalizability purposes.

What do you mean by f and p weights? Does is it possibly mean personal-level weights and household- representative weights?

According to my understanding, personal-level weights should total to the final population number at the time the survey was taken. Personal-level weights would be the weights that you would be using if you are answering a question on a specific personal-level such as the number of diabetics in the study population (since the sample was randomly selected and you have a number of people who reported to be diabetic).

Unfortunately, I do not have much experience in using household-representative weights, but I assume you would be using the household weights if you were answering a question on a household level, such as of food-insecurity reported by an individual in the household. Because if you use person-level weights for food insecurity among households for example, and if that person-level weights variable represented adults per-say, the variable will reflect the number of adults ONLY among the household and will overlook children since person-level weights is only capturing adults in that survey.

It is essential that you review the weighting steps that were followed in order to decide exactly which weighting variable you will be using in your analyses.

I hope that this was helpful, please do not hesitate to ask any further questions.

And I agree with the previous answers, weights do not necessarily have to be integers.

Best wishes,

Razan