|
<< Click to Display Table of Contents >> Navigation: Reference Manual > Windows and dialog boxes > Weight data > Weight data window |
From the Main window menu [Data] [Weight data file] is used to calculate respondent weights from targets. It has a menu.
This window is used to open a data file and calculate the respondent weights needed to achieve the targets set with the Weight targets window.
For a description of weighting see User Guide, Weighting introduction
The targets can be simple grossing up or down, full matrix, rims, accumulated quantity targets, or any combination of these . Entries used for targets can be any single-coded, multi-coded, integer or float questions or variables, and filters.
Weighting can be applied to subsets of the data.
The targets set can be stored in a weighting file for re-use later.
The calculated weights can be stored in the data file (in a float question) and they can be output to a CSV file.
Various statistics including the effective sample sizes (ESS or efficiency) are shown during the weighting procedure and a report can be produced with the final details.
You can set the precision required which will reduce or increase the time taken to weight the data.
The precision is set as the number of significant places that are required to be the same as the target required. For example if your weighted target total is 1000 then you will probably want 4 significant places to ensure that the base on the tables is what you expect. If your total is 10000 then 5 places are needed.
The weighting procedure adjusts the weights for individual respondents using a simple formula. This results in different records getting different weights.
You can constrain the weight applied to each record to a range of values. For example if you set a range of 0.5 to 2.0 then no record will be given a weight of less than 0.5 or more than 2.0.
At the end of each iteration the window shows the number of records that have reached the limits set.
Setting too small a range may result in the program not being able to achieve the targets set.
You can choose to stop the calculations when one of the constraints is met.
The weights are calculated as follows:
•Open the data file
•Set the targets with these choices:
oWith the Weight targets window
oLoad an existing Weighting file and change if required
•Create or select an existing float question to hold the calculated weights
•Prepare and check the data with these choices:
oStart with all weights set equal
oStart with existing weights already in the data
•Calculate weights:
oYou can repeat this step without resetting the weights beforehand if required
oYou can request a single iteration to see the effect of one pass over the data
•Amend the calculated weights to show them as they will be stored in the data, there may be some loss of precision if insufficient data locations have been allocated to the weight question
•Save the targets to a Weighting file in case you need to recalculate weights because the data has changed
•Save a summary report of the weighting statistics
•Save the data file with the individual weights added
•Save the project file if you have created a new float question to store the weights
When this window is closed you may see Weight save with a reminder of any information that has not yet been saved.
You can stop the weighting procedure at any time. It will stop at the end of the current iteration which may take a few seconds.
You can alter the settings for precision and constraints before continuing with the iterations. You can use the "continue" button or the menu to carry on. Using the menu you can request a single iteration to see check what the changes are.
Calculation iterations only will stop when one of the following occurs:
•Every target is correct to the number of decimal places set
•You use the [Stop] button
•Neither the average nor the worst significant digits have improved for a number of consecutive iterations
The window has three areas:
This area shows the results of the calculations so far for each target set:
•Name of the entry being used
•Entry response or base being used
•Type of target set
•Actual number of respondents in the data
•Usable respondents in the data (those without a fixed, predetermined weight)
•Target set
•Progress: reaches 100% when this target has been achieved
•The weighted figure achieved
•The effective sample size (ESS)
•Efficiency: shows loss of data due to weighting effects (100% = no data lost)
•Minimum weight used
•Average weight used
•Maximum weight used
•Description of this target
This shows the respondent data being used for the weighting including the calculated weights and the target entries.
Here you can choose the precision required for the calculated weights which is set as significant digits. The greater the significant digits, the longer the calculations will take. The maximum setting is 14 significant digits because this is the limit that can be achieved using 64 bit floating point numbers. This area also shows the average target significant digits and worst significant digits achieved for any target set.
You can also set the range of values permitted for the individual weights (weighting factors). No record will be given a respondent weight outside the range set. The number of records whose weights has reached the limits set is shown alongside.