Why survey data usually needs weighting
Survey data almost never mirrors the population it came from. Perhaps women make up 60% of your sample but only 51% of the population, or younger respondents were oversampled on purpose so you would have enough of them to analyse. If you analyse the raw data as it stands, every percentage and every mean is tilted towards the groups that are over-represented.
Weighting cases in SPSS corrects that tilt. You give each respondent a weight — a number that says how many people in the population that one case stands for — and SPSS counts over-represented cases a little less and under-represented cases a little more. The result is estimates that describe the population you actually want to talk about, not just the people who happened to answer.
This tutorial shows you how to weight cases in SPSS for survey data: when weighting is appropriate, how to set it up click by click, the syntax you need, how to check that it worked, and how to report weighted results in APA style.
What does weighting cases mean in SPSS?
When you weight cases in SPSS, you tell the program to treat each row of your data file as if it occurred a certain number of times. A case with a weight of 2 counts twice in frequency tables, means and crosstabs; a case with a weight of 0.5 counts half as much as a case with a weight of 1.
SPSS applies weights as frequency weights. That is an important detail, because it affects how you should think about the output. SPSS does not know that your weights came from a complex sampling design — it simply repeats each case according to its weight. Weighted counts and percentages are therefore correct, but standard errors and significance tests are calculated as if the weighted sample size were real. If your weights add up to much more than your actual number of respondents, your tests will look more precise than they really are. For design-based inference with survey weights, the Complex Samples module is the technically correct tool; for descriptive reporting and most thesis work, standard case weighting is what supervisors expect and what this tutorial covers.
When should you weight cases?
Weighting is appropriate whenever the people in your sample do not represent the population in the proportions you want to generalise to. The most common situations are:
- Oversampling. You deliberately sampled extra respondents from a small subgroup — for example, extra interviews in a rural region — so you could analyse that group separately. Weighting scales each group back to its true share of the population.
- Stratified sampling. Your design sampled different strata at different rates. Weights restore each stratum to its population proportion.
- Post-stratification. Your sample happened to come out unbalanced on a known characteristic, such as age or gender, and you know the true population figures from census or registry data. Weights adjust the sample to match.
- Non-response adjustment. One group responded at a much lower rate than others, and you have a defensible basis for upweighting the respondents you did get.
Do not weight cases just because the option exists. If your sample is a simple random sample and you have no population figures to calibrate against, weighting adds complexity without improving anything. Never weight by a variable you chose after seeing the results, and never weight to make an inconvenient finding disappear — that is not adjustment, it is distortion.
Before you start: your weight variable
SPSS weights cases using a variable already in your data file, so the first step is making sure that variable exists and is correct.
The weight variable must be numeric. Each value is the weight for that case. A value of 1 means the case counts exactly once; 2 means it counts twice; 0.5 means it counts half as much. SPSS ignores cases with a weight of zero, a negative weight, or a missing value on the weight variable, so check for those before you proceed.
Many surveys supply the weight variable for you. Large public datasets usually include a weight variable with a name like WEIGHT, WGT or FINALWGT. If your data came from a survey agency, look in the codebook first — do not invent your own weights when official ones exist.
If you need to create weights yourself, the usual formula is simple: weight equals the population share of a group divided by the sample share of that group. Suppose women are 51% of the population but 60% of your sample. Each woman’s weight is 51 / 60 = 0.85, and each man’s weight is 49 / 40 = 1.225. Create the variable with Transform > Compute Variable, using an IF condition for each group, or compute it in a spreadsheet and merge it in.
Normalise your weights if you can. Weights that average 1 keep the total weighted sample size equal to your actual number of respondents, which keeps significance tests roughly honest. To normalise, divide each weight by the mean weight: Transform > Compute Variable, with the expression weight / mean-weight. If your weights already average close to 1, you can skip this step.
How to weight cases in SPSS: click-by-click steps
With a numeric weight variable in your file, weighting takes less than a minute:
- Go to Data > Weight Cases.
- Select Weight cases by.
- Move your weight variable into the Frequency Variable box.
- Click OK.
That is all there is to it. Look at the status bar at the bottom of the SPSS Data Editor: it now reads Weight On. Every analysis you run from this point forward — frequencies, descriptives, crosstabs, t-tests, regression — will use the weights until you switch them off.
To turn weighting off, go back to Data > Weight Cases, select Do not weight cases, and click OK. The status bar changes back to Weight Off. Make a habit of switching weighting off when you move on to unweighted work: forgetting that weights are still active is one of the most common sources of confusing output.
SPSS syntax for weighting cases
If you prefer syntax, or you want your weighting step saved in a syntax file so your analysis is reproducible, the commands are short:
WEIGHT BY weightvar. FREQUENCIES VARIABLES=gender age_group /ORDER=ANALYSIS.
Replace weightvar with the name of your weight variable. Run the WEIGHT BY command once and it stays in effect for every procedure that follows in the same session. To switch weighting off in syntax, use:
WEIGHT OFF.
A good habit is to put the WEIGHT BY command at the top of your analysis syntax file with a comment line explaining what the weights are and where they came from, so anyone reading your syntax — including your future self — can see exactly what was weighted and why.
How to check that weighting worked
Before you trust any weighted result, verify the weighting with a quick frequency table:
- Run Analyze > Descriptive Statistics > Frequencies on a variable whose population distribution you know, such as gender or age group.
- Compare the percentages with the known population figures. After weighting, they should match closely.
- Check the total N. If your weights are normalised, the weighted total should equal your actual sample size. If it is much larger or smaller, your weights are not normalised — which is fine for percentages, but remember it when you look at significance tests.
- Run the same frequency table with weighting switched off and compare. The unweighted percentages show your sample; the weighted percentages show your adjusted estimates. If they are identical, either your sample was already balanced or the weighting did not take effect — check the status bar.
Also scan for surprises: a single case with an extreme weight can dominate your results. If one respondent carries a weight of 40 while everyone else carries a weight near 1, find out why before you report anything.
How to interpret weighted output
Weighted output looks like ordinary SPSS output, but a few things deserve attention:
Counts are weighted counts, not people. A frequency table may show 512.4 respondents in a category. The decimal is normal — it means the weighted count is not a whole number of people. Report the percentage, and if you report the count, round it and make clear it is a weighted count.
Percentages are the point. The main thing weighting changes is the percentages and means. Compare weighted and unweighted results side by side in your write-up if the difference matters for your argument; readers trust a weighting decision more when they can see what it changed.
Significance tests need a caveat. Because SPSS treats weights as frequency weights, a weighted sample size larger than your real sample size makes standard errors smaller and p-values more impressive than they should be. With normalised weights this distortion is small. If your weights are far from normalised, either normalise them or add a sentence in your methods noting that standard errors from weighted analyses should be interpreted cautiously.
Missing data still matters. Weighting does not fix missing data. A case with a missing value on the variable you are analysing is still excluded from that analysis, weight or no weight. If non-response is your problem, deal with it as a missing-data issue first, and only then weight.
Common mistakes when weighting cases in SPSS
- Weighting by the wrong variable. A grouping variable such as gender coded 1 and 2 is numeric, so SPSS will accept it — and your men will count twice as much as your women. Always double-check that the variable in the Frequency Variable box is your actual weight variable.
- Leaving weights on. Weighting stays active until you switch it off, across analyses and across data files opened in the same session. Check the status bar before every new analysis.
- Using weights that sum to the population size. If each weight represents thousands of people, your weighted N will be in the millions and every test will be “significant”. Normalise weights to average 1 unless you have a specific reason not to.
- Weighting to fix a bad sample. Weights adjust proportions; they cannot create information about people you never sampled. If an entire subgroup is missing from your data, no weight can bring it back.
- Not documenting the weights. Your methods section must say where the weights came from, how they were calculated, and whether they were normalised. A weighted analysis with no explanation looks like a manipulation, even when it is perfectly legitimate.
Reporting in APA style (7th ed.)
APA style has no single prescribed format for weighted analyses, but the principles are clear: describe what you did in the Method section and label weighted figures wherever they appear.
In your Method section, under Participants or Data Analysis, write a sentence such as: “Analyses used sampling weights to adjust the sample to population distributions for age and gender. Weights were normalised to a mean of 1.0, so the weighted sample size equals the unweighted sample size (N = 412).”
In tables, add a note beneath any weighted table: “Note. Frequencies are weighted; percentages are based on weighted counts.” Report the unweighted sample size alongside the weighted figures so readers know how many real people the estimates rest on — for example, “52% (weighted n = 214; unweighted N = 412)”.
In text, a weighted result reads like any other: “After weighting, 52% of respondents reported using public transport daily, compared with 47% in the unweighted sample.” The difference between the two figures tells the reader exactly what the weighting did.
Key Takeaways
- Weighting cases in SPSS adjusts your sample to match known population proportions — use it when your sample is unbalanced on a characteristic you can measure against population figures.
- Your weight variable must be numeric, with no zero, negative or missing values for cases you want included.
- Set it up in seconds via Data > Weight Cases, or with the syntax WEIGHT BY weightvar. Switch it off with WEIGHT OFF when you are done.
- Normalise weights to an average of 1 so the weighted sample size matches your real sample size and significance tests stay honest.
- Always check weighting with a frequency table, report weighted percentages clearly, and document where the weights came from in your Method section.
Frequently asked questions
What does “Weight On” mean in SPSS?
It means case weighting is currently active: every analysis you run will count each case according to its value on the weight variable. You can see it in the status bar at the bottom of the Data Editor. To switch it off, go to Data > Weight Cases and select “Do not weight cases”.
Can I weight cases by a string variable?
No. The weight variable must be numeric, because SPSS uses its values as multipliers. If your weights are stored as text, convert them with Transform > Automatic Recode or by re-entering them as numbers first.
Why are my weighted frequencies showing decimals?
Because weighted counts are sums of weight values, not counts of people. A category total of 214.6 means the cases in that category carry weights adding up to 214.6. Report the percentages, and round weighted counts if you quote them.
Do weights affect every SPSS procedure?
Almost all of them — frequencies, descriptives, crosstabs, t-tests, ANOVA, regression and correlations all respect case weights. A few specialised procedures ignore weights or require the Complex Samples module for design-correct standard errors, so check the documentation if a procedure’s output looks unweighted.
Should I weight my data if I used convenience sampling?
Only if you have reliable population figures to weight against and a defensible reason to believe the weighting corrects a known imbalance. Weighting cannot turn a convenience sample into a probability sample, and it cannot fix the absence of whole groups. When in doubt, report unweighted results and discuss the sample’s limitations honestly.
Need this done on your data? Get a free quote → /services/