Assigning survey weights

[Mishel Shahid]

Hi guys. I am working with a longitudinal data set and I am supposed to assign weights or adjust for weighting before analysing the data. Longitudinal survey weights have been provided as a variable in the data set. I was told that I am supposed to use the svy command for this in stata but have no idea how this works. Any help or suggestions will be greatly appreciated.

[Joro Kolev]

Hard to say anything in particular with the minimal information you provide.

Check out

help svyset

[Mishel Shahid]

Hi Joro, thankyou for your reply.

To clarify I have a national logintudinal data set that I have to work with. In this data there are variables from two time points 10 years apart. A variable for the strata from both time points was provided with a longitudinal weight varible. I then survey set the data by using the participants as the PSUs as this was a cluster sample. I set it in two stages using the strata provided and assigned the survey weights provided in the dataset and used a linearized VCE. On setting using the IDs as PSUs and Strata from first and second time point it was prompted that the second stage is automatically ignored to estimate the variance so ultimately the data was set using the first stage and its strata only. I am unsure if this is appropriate because i am having the following issue. When i create a one way table for the distribution of males and females in the sample I get the following result:

Number of strata = 10 Number of obs = 1,598
Number of PSUs = 1,598 Population size = 14,164,014
Design df = 1,583

----------------------------------------------------------
Sex |
| percentage lb ub obs
----------+-----------------------------------------------
male | 50.14 47.04 53.24 595
female | 49.86 46.76 52.96 1003
|
Total | 100 1598
----------------------------------------------------------

As you can see, the observations in female are higher but for some reason the percentage is lower than males.

Sex | Freq. Percent Cum.
------------+-----------------------------------
male | 598 40.89 40.89
female | 1,009 59.11 100.00
------------+-----------------------------------
Total | 1,607 100.00

Now when i run the same table without the svy analysis it shows the expected results of the females being more in number and percentage reflecting that.
I am unsure why this descripancy is there. So far the gender is the only category that has shown this kind of result.
Could it be because the survey set was done incorrectly? Why would the perecentage of females (49.86) be showing less than males (50.14) while their frequencies are the opossite females (1003) males (595).
I appreciate any help or guidance as always. Hope my query is clear this time and avoids any confusion.

[David Radwin]

Could it be because the survey set was done incorrectly?

Maybe. Please read the documentation that accompanies the dataset to see if it includes the Stata command to correctly svyset the dataset. Then try to replicate some statistics published by a reliable author (such as a government agency or a peer-reviewed publication), including the corresponding standard errors or some other measure of variance. If your results are not nearly identical, you are probably doing something wrong or at least different from what the data provider intended.

Why would the perecentage of females (49.86) be showing less than males (50.14) while their frequencies are the opossite females (1003) males (595).

If I understand correctly that the first table is weighted and the second table is unweighted, it means that the percentage of females in the sample was higher than the percentage of females in the population, so the survey weights adjust the percentage of females in the sample to reflect the percentage of females in the population. This discrepancy between the sample and the population might be due to a higher response rate from females, greater attrition by males, deliberate oversampling of females, or any number of other reasons. The survey weights are intended to make the sample equivalent to the population.

[Mishel Shahid]

Thankyou David.
Your explanation of why the discrepancy could be there because of using the survey weights makes alot of sense and I am thinking its possibly that. As no other variable in my initial svyset descriptive analysis has shown a similar trend. Its just gender that is showing this issue.
I really appreciate your help. It puts things into perspective.