SPSS Different Sample Size in Output: Why Your Tables Don’t Match

If you have ever noticed SPSS different sample size in output across your Descriptives, Correlations, and Regression tables, you are not losing your mind, and you have not broken your dataset. This also shows up as SPSS different N in different analyses the moment you start comparing tables side by side, and it is one of the most common questions I get from dissertation students in my twelve years of doing this work, usually sent to me at 1 a.m. with three screenshots and a lot of panic in the message.

I am going to walk you through exactly why this happens, which number you should actually report, and when a dropped sample size is a genuine problem versus completely normal statistical behaviour.

The Short Answer: Why SPSS Shows a Different Sample Size in Output

Here is the direct answer before we go further. SPSS shows different N in different analyses because each procedure decides, on its own, which cases to drop when there is a missing value involved. A missing age in one row might not matter for your Descriptives table but might knock that entire case out of your Regression table.

It’s Not a Mistake. It’s How SPSS Handles Missing Data

Every time you run a test, SPSS checks the variables involved in that specific test. If any variable in that test has a missing value for a given respondent, SPSS has to decide whether to drop that respondent, keep them, or drop them only for the pairs where data is missing. That decision rule is different for different procedures, and that is the entire reason behind the mismatch.

The Three Numbers You’ll See: Total N, Valid N, and Analysis N

• Total N is your full dataset. Everyone you collected a response from.
• Valid N is what SPSS says survived for that specific variable or table.
• Analysis N is the number that actually went into a test like regression or ANOVA, after every variable in that model was checked for missing data.

If you have noticed SPSS valid N different from one table to the next, this is exactly where that number is coming from. These three numbers are rarely identical once you have more than a handful of variables, and that is expected, not an error.

Case Processing Summary: Where SPSS Tells You the Real Story

Most students skip straight past this table because it looks boring. It is actually the most honest table in your entire output.

How to Read the “Valid N (listwise)” Footnote

Valid N (listwise) is the number of cases that had complete, non-missing data on every single variable included in that particular analysis. If your footnote says “Valid N (listwise) = 189,” it means 189 respondents had no missing values anywhere across the variables you entered into that model.

Finding Out Exactly How Many Cases Were Excluded

SPSS actually has a dedicated procedure for this, called Missing Value Analysis (Analyze, then Missing Value Analysis). It gives you a full missing-data report in one place, instead of piecing it together manually. I usually start here before touching anything else, because it is built exactly for this job and most tutorials skip straight past it.

If you prefer a simpler route, go to Analyze, then Descriptive Statistics, then Frequencies, and tick the box for missing values. SPSS will give you a clean count of how many cases were excluded and from which variable. I make every one of my students run one of these two checks before they touch a regression model, because guessing is how dissertations get flagged at viva.

Listwise Deletion vs Pairwise Deletion

This is the part that actually explains SPSS cases excluded from analysis, and it is where most generic tutorials stop one sentence too early.

How Listwise Deletion Works in SPSS

Listwise deletion drops a respondent from the entire analysis if they are missing even one value across any of the variables involved. One missing answer on a 40-item questionnaire is enough to lose that whole case from a regression model that uses all 40 items.

How Pairwise Deletion Works Instead

Pairwise deletion only drops a case for the specific pair of variables where data is missing, not for the whole analysis. This is a big reason correlation sample size differs in SPSS so often: a correlation matrix with ten variables might show ten different N values, one for each pair, because SPSS calculates each correlation using only the respondents who answered both variables in that pair.

Which One SPSS Uses by Default in Each Procedure

1. Regression defaults to listwise deletion.
2. ANOVA and MANOVA default to listwise deletion.
3. Bivariate Correlations defaults to pairwise deletion.
4. Factor Analysis defaults to listwise deletion unless you change it manually.

IBM’s own SPSS documentation confirms these defaults clearly, but it stops at the mechanism. Most SPSS tutorials copy this same list and stop there too, as if knowing the default rule is the same as knowing whether that default is right for your dataset. It isn’t. That judgment call, the one that actually matters for your results chapter, is usually the part missing from generic guides.

Why Your Regression N Is Always Smaller Than Your Dataset

This is the question I get asked the most, phrased in every possible way, and the answer to why your regression N is smaller than the dataset SPSS shows is always the same mechanism.

More Predictors Mean More Chances to Lose a Case

A simple correlation only checks two variables for missing data. A regression model with six predictors checks all six, plus the outcome variable, for every single respondent. More variables in the model means more opportunities for one missing value to knock a case out. A smaller analysis N also means fewer degrees of freedom for your model, which directly affects how easily your effect sizes reach statistical significance.

A Real Example: 250 Responses to 189 in the Regression Table

I worked with a master’s student who had 250 completed survey responses and was convinced her dataset had been corrupted when her regression output showed an N of 189. She had already read three different blog posts about this before messaging me, and every one of them explained listwise deletion in the abstract without ever telling her how to check whether her own drop of 61 cases was dangerous or harmless. That is exactly the gap I want to close here.

Nothing was wrong with her file. Her outcome variable plus five predictors meant that any respondent missing even one answer across those six variables was dropped under listwise deletion, and once you add up small missing-data rates across six separate variables, losing 61 cases is entirely plausible.

Is This a Problem? My Honest Take After 12 Years of Reviewing Dissertation Data

Most SPSS guides treat every dropped case as equally fine or equally alarming. In my experience, it depends entirely on how much you lost and why.

When a Dropped N Is Completely Normal

A 5 to 10 percent drop from total N to analysis N, with no obvious pattern in who got excluded, is routine and rarely worth a committee’s attention.

When It’s a Red Flag You Need to Investigate

If you are losing 20, 30, or 40 percent of your sample, or if the people being dropped share something in common, such as all being from one age group or one survey wave, pay close attention. That pattern of missing values reducing your sample size can bias your results.

Statisticians describe missing data using three patterns. MCAR (missing completely at random) means the missingness has nothing to do with any variable in your study. MAR (missing at random) means it depends on other variables you measured, but not on the missing value itself. MNAR (missing not at random) means the missingness depends on the value that is actually missing. This is the dangerous one, because it can quietly bias your results without any obvious warning sign.

This exact scenario comes up constantly in SPSS missing cases dissertation analysis work, and it is almost always a sample size conversation before it ever becomes a statistics conversation.

The 5 to 10 Percent Rule I Use With My Own Students

Anything under 10 percent total case loss, I generally tell students to proceed and simply report it honestly. Anything above that, I ask them to run Little’s MCAR test on the missingness pattern before writing a single word of their results chapter, specifically to check whether it looks like MCAR or leans toward MNAR.

Which Sample Size Do You Actually Report in Your Methods Chapter?

This is the commercial heart of the whole question, because it decides what goes on paper in front of your examiner.

APA-Style Reporting Templates You Can Copy

• For your overall sample: “A total of 250 participants completed the survey (N = 250).”
• For a specific regression model: “Due to missing data, the regression analysis was based on a valid sample of 189 participants (N = 189).”
• For a correlation with varying pairwise N: “Sample sizes for individual correlations ranged from 201 to 243 due to pairwise deletion of missing values.”

Report the N that was actually used for that specific test, next to that specific table, every time. Do not report one single N for your whole results chapter and hope nobody checks the footnotes, because examiners do check the footnotes.

How to Explain N Differences to Your Supervisor or Examiner

One short sentence in your methodology section, stating which deletion method SPSS used and why, closes this question before it ever becomes a viva question. I have seen students lose confidence in their own viva purely because they could not explain a number that was sitting right there in their own appendix.

Fixing It: Your Options Beyond Just Accepting the Smaller N

You are not stuck with whatever number SPSS hands you by default.

Step-by-Step: Diagnosing and Handling Missing Data Before You Panic

1. Run Missing Value Analysis (or Frequencies with missing value counts) to see exactly which variables are affected.
2. Check whether the missingness looks random or whether it clusters around a specific group.
3. Decide, variable by variable, whether listwise or pairwise deletion actually fits your analysis.
4. If missingness is non-random and substantial, consider running a proper multiple imputation model rather than deleting cases outright.
5. Recalculate your statistical power using your actual analysis N, using the sample size calculator, before you interpret significance.

Multiple Imputation as a Presentable Alternative

Rubin’s (1987) multiple imputation framework, laid out in his book Multiple Imputation for Nonresponse in Surveys, lets you estimate plausible values for missing data instead of deleting cases outright, which keeps your full sample intact for analysis. Plenty of guides mention this option as if ticking a box in SPSS is the whole job done. It isn’t.

The method needs a defensible justification in your methodology chapter: why you chose it, how many imputations you ran, and which assumption about your missing data pattern you are relying on. Skip that explanation and an examiner will ask for it anyway.

Recalculating Power After Case Loss

A sample that was adequately powered at 250 may no longer be adequately powered at 189. I ask every student to rerun a power check in G*Power once they know their real analysis N, particularly before they write any sentence claiming a non-significant result is meaningful.

Common Mistakes Dissertation Students Make With Missing Data

• Manually deleting rows in the data view before running any test, which hides the real reason your N changed.
• Reporting a single overall N in Chapter 3 that does not match any specific table in Chapter 4.
• Ignoring table footnotes entirely, then being unable to answer a direct viva question about them.

Correlation output is a particularly common place this goes wrong. If you want the deeper mechanics of how correlation choice itself affects your results, my guide on Pearson vs Spearman correlation walks through it alongside this exact N issue. Regression students chasing the same problem from the diagnostics side should also look at my Durbin-Watson test interpretation guide, since autocorrelation checks run into the same sample size footnote.

If this entire area of SPSS output still feels unfamiliar, my full SPSS tutor guide covers the basics of reading every output table SPSS generates, not just this one issue.

FAQ

About the author: I am Siddharth, and I have spent the last 12 years helping dissertation and thesis students make sense of exactly this kind of SPSS output, across the UK, US, and UAE. If your own case processing summary does not match what is described here, get help interpreting your own SPSS output.