Dear All,

First off, apologies for the long post.

I am having a challenging with how to go about with svysetting my data. within my data set I have popweight (population weight) and panweight (panel weight). The survey is a 2017 survey but contains hh that were interviewed in 2014 (old and new households). I am interested in both the hh from 2014 and 2017 who sold crops between 2016 and 2017 (within the dataset, there is a variable that asks whether these households sold any crops between 2016 and 2017).

The only additional information I have on the survey on sampling is from the report and from this I have concluded that my psu = cluster, the appropriate weight I should use = panweight, and strata = province
hence I should svyset my data as follows

gen strata = province
gen psu = cluster
gen wt = panweight (using panweight and not popweight because I am interested in both 2014 and 2017 hh)

svyset psu [pw=wt], strata (strata)

Is this correct. Below is what the report says on the sampling and weights

The 2017 survey is a panel survey continuing from 2014 survey. The sampling frame was based on the 2010 Census of Housing and Population. A stratified two-stage sample design was used for the 2014 sampling. The first stage involved identifying the Primary Sampling Unit (PSU). The PSU was defined as one or more Standard Enumeration Areas (SEAs) with a minimum of 30 agricultural households. At the second stage, all households in selected SEAs were listed and agricultural households identified. Listed agricultural households were then stratified into three categories, A, B, and C, on the basis of total area under crops; presence of some specified special crops. Systematic sampling was then used to select 20 households distributed across the three strata in each SEA.
The 2014 survey covered 442 Standard Enumeration Areas (SEAs) across the 10 provinces and a total of 8,840 households. With the 2017 survey an additional 34 SEAs were added. This brought the total households to be interviewed to 9,520. This sample was expected to yield reliable estimates at provincial and national levels.

Sample weights
The first survey was conducted in 2014 and a follow-up survey in 2017. Two sets of weights were generated from the survey data, panel weights and the population weights. Panel weights are for analysis that utilize both 2014 and 2017 survey, while the population weights should be used for standalone cross-sectional surveys.

Panel weights
The panel weights were calculated by adjusting the final weights calculated for the 2014 survey, with the survey response information obtained for each of the three farm categories (Category A, B and C) as described above. The adjustment factor for non-response was calculated by dividing the number of households selected during the 2014 survey by the total number of households responding during the 2017 survey in each of the three farm categories.
The adjustment factors based on projected number of agricultural households were calculated by dividing the projected number of agricultural households in 2014 by the projected number of agricultural households for 2017. The adjustment factors were then multiplied with the preliminary weights to obtain the final weight.

cross-sectional weights
The 2017 survey was a combination of two samples, 2014 survey sample and 34 additional clusters from three other provinces. The additional clusters were added to the 2017 survey in order to have statistically valid estimates for natural resources management issues in these Provinces.
In order to combine the two samples, the probabilities of selection of the original sample for all the strata districts identified with new clusters were re-calculated as additional Standard Enumeration Areas were added. The general procedure for calculating the weights made use of sampling probabilities at first- stage selection of SEAs and probabilities of selecting the households. The weights of the sample are equal to the inverse of the probability of selection.


Thanks