Setting targets for weighting data (respondent weighting)
See Weighting respondents for an overview of rim/target weighting.
Targets described
A target is the weighted total that you desire for a particular subset (entry response) of the data. It can expressed as:
- The required total after weighting
- The required percentage of the total after weighting.
- If you filter the entry, then it can be the required percentage of those who pass the filter
- Keep the weighted total the same as the un-weighted total
The QPSMR Companion program automatically adds weighting targets for whole data file first. The program sets the weighted total the same as the un-weighted total. You can change this to a required total which need not be the same as the un-weighted total.
For each subset of the data you can instead choose to:
- Ignore this response, meaning do not attach a target to this response
- Preserve all weights already assigned to records with this response
- Apply a zero weight; however, do not include these records in the weighted analysis
- Apply a set fixed weight to the subset, every record gets the same weight.
You will not use the above options very often. If a particular response to a question not used as a rim is very important, it might be worth giving it a fixed weight so that no weighting effects will apply to these particular records.
Weighting Targets entries
Single-coded
Each questionnaire will be in only one of the responses. You will usually set a target for every response. It is important that there are no missing values for single coded entries, because the targets will normally need to add to 100% of the total.
Multi-coded
Each questionnaire will be in none, one, or more than one, of the responses. You can assign targets to one or more responses to a multi-coded entry. If you use all responses, then the targets should add to more than 100%.
If there are a lot of responses it will be better to only assign targets to the important responses for the survey.
Combined multi-coded
For convenience you can combine many single-coded and multi-coded entries into one variable for weighting. The rules for assigning targets will be same as if they were separate entries.
Quantity
You will rarely use these, the other targets must all be percentages, and you will need to “Ignore” the first row (all records). You can use this target to gross up or down the whole data file to make the total for the quantity.
The Companion program does not apply variable weights depending upon which side of the mean each record lies.
Filtered entries
If you filter an entry (not asked of everyone), then Companion will create two addition rows: one for all those who passed the filter (who saw the question), and one for those who did not pass the filter. You can ignore these extra rows or set target targets for them as normal.
If you set percentage targets you can choose whether the target is the percentage of all records, or only of those who passed the filter.
Ignored entries
You can add entries and leave all the responses set to “Ignore”. You might want to do this to check the “weighting effects” on entries that you will not actually use for the weighting.
How to set weighting targets
It is very easy to set targets. Simply tell Companion what weighted totals or percentages you require and the program will do its best to achieve them:
From the Weight data window select [Targets] [Targets set] to open the Weight targets window. The subsets to be given targets need to be the responses to one or more entries in the project.
Use [Entries] [Add an entry to the list] to add each entry (rim) needed for the weighting. You can then select one or more responses from the list to allocate the type and value of target you would like to apply.
If you have a list of targets in a CSV or spreadsheet file you can highlight a number of responses and paste the targets for all those selected at once.
Use the first row {all targets} to set the weighted total for all records in the data file. You can usually set this to “As un-weighted” so that the weighted total will be the same as the un-weighted total (the number of records in the data file). It is important you use this target to “tie down” the total if percentage targets are used.
Rim targets
The most common form of target is a rim. This is a single-coded entry where every record is in one of the responses. You usually set a percentage target for each response. These percentages must add to exactly 100%. If they add to 99.9% or 100.1% then the weighting calculations cannot succeed. You can check the sum of the percentages by selecting all the responses for the rim and the total percentage will be shown at the bottom of the window.
It is important that there is no missing data for the rims. If there is missing data then these records will normally be excluded from the survey. If this is not practical then a target will need to be allocated for these records and the percentages need to add to less than 100% leaving the remaining percentage for the missing data.
It is also important that there are records in each response to a rim. If there are no records with a particular response then it is impossible to reach the target because there are no records to weight. Use “Ignore” for any responses where there is no data.
You can have as many rims as you wish, although the more you use, the lower the ESS (effective sample size) will be because you will increase the weighting effects.
Matrix targets
Sometimes the targets will be a full matrix. For example if the targets are a grid of Age (3 categories) by Area (4 categories) then there will be 12 targets. Treat this matrix as an ordinary rim with 12 responses. You can create the rim variable responses in Companion using [Responses] [Add response matrix].
Multi-coded
Sometimes it is necessary to allocate targets for a multi-coded entry, for example brands purchased in last month. The targets for such a question will add to more than 100%.
It is not usually practical to allocate a target to every response. For a long list of responses it is recommended that targets are only set for the few brands that are important to the survey. Use “Ignore” for the other brands. It may be necessary to start with one or two responses with targets and add more only if the ESS (effective sample size) is not adversely affected.
Partial targets
It is not necessary to allocate targets to all responses in target entries. In some cases it may only be appropriate to allocate targets to one or two responses and ignore the rest. This avoids unnecessary adjusting of weights for records with responses for which the weighted total is not important.
Mixed targets
You can mix together any of the above types of targets. The only point to remember is to tell Companion the weighted total that you need for each target. You do not need to concern yourself with the technical details of how the program will achieve this.
Ignored targets
You can add entries and leave all the responses set to “Ignore”. The weighted totals for these entries will be shown in the weight summary. This can be useful to check the effect of the weighting on entries not actually used as targets.
Efficiency priority
You can control the priority with which each of the targets is calculated. A higher figure (above 1.0) means that this target has a higher priority, and this target will be attempted more strongly that the other targets. This will increase the ESS and efficiency for these records. This could be done for important responses, for example the clients own brand. A lower figure (under 1.0) means that this target has a lower priority and this target will be attempted less strongly that the other targets. This may reduce the ESS and efficiency for these records. This could be done when this target is not very important.
Saving targets
After you have closed the Weight targets window to return to the Weight data window you can use [Targets] [Save to weighting file] to save the targets in an XML file. If you need to do the weighting again you can use [Targets] [Load from weighting file] to retrieve them. If you have not saved the targets when you close the Weight data window you will be prompted to do so.
Before you save the targets you should choose the question that the weights are to be stored in.