The data preparation decisions that quietly corrupt results, why syntax matters more than menus, and what the output will never warn you about.
SPSS is a capable and widely taught statistics package, and its main risk is exactly that: it will run whatever you ask it to, and it never objects. Nothing warns you that a variable is set to the wrong measurement level, that your missing value code is being treated as real data, or that the test you chose does not suit your design. Most SPSS problems in doctoral work originate before any analysis is run, in how the data was set up.
Time spent getting the dataset right is repaid several times over. Three settings cause most of the trouble.
Measurement level. Each variable is defined as scale, ordinal or nominal, and this affects which procedures are offered and how some output behaves. Imported data frequently arrives with everything defaulted to one setting, and nothing prompts you to check. Set it deliberately for every variable before starting.
Missing value codes. The single most damaging setup error. If you coded “prefer not to say” as 99 and did not declare 99 as a missing value, SPSS treats it as the number ninety-nine — and every mean, correlation and regression involving that variable is wrong. The output looks entirely normal. Declare missing codes explicitly, then check your descriptives: a variable on a five-point scale with a mean above five has told you something is wrong.
Value labels. Recording that 1 means male, 2 means female, and so on. Skipping this produces output you cannot interpret six months later, and tables you have to annotate by hand for the thesis.
One habit worth adopting immediately: keep the raw file untouched and work on a copy. Every recode, exclusion and computed variable happens on the working file, so you can always return to the original. “I cannot remember what I did to this variable” is a genuinely serious problem at write-up, and it is entirely avoidable.
Most people learn SPSS through its menus, and everything can be done that way. Using syntax instead — or as well — solves a problem that will otherwise surface at the worst possible moment.
Menu clicks leave no record. Six months later, writing up, you will not remember whether you excluded cases before or after computing a scale, which of three recodes produced the variable you analysed, or exactly how you handled that awkward item. Syntax is a written record of every step, and it can be re-run from the raw data to reproduce your results exactly.
That matters for three concrete reasons. Your methodology chapter has to describe what you did, and syntax is the record. If a supervisor or examiner questions a result, you can re-run it and show the working. And when you discover an error in an early step — which happens — you can fix it and re-run everything after it in seconds rather than repeating an afternoon of clicking.
You do not need to learn to write syntax from scratch. Every dialog box has a button that pastes the equivalent syntax into a file instead of running it directly. Use the menus, paste rather than run, and you accumulate a complete record with no extra effort. Keep that file, comment it, and it becomes an appendix.
Running the analysis is the fast part. These steps are what make the result trustworthy, and skipping them is what produces findings that fall apart under questioning.
SPSS produces generous output, much of which is not what you need, and some of which is easy to misread.
Points that recur in doctoral work:
Do not paste raw SPSS tables into the thesis. Rebuild them in your discipline’s reporting format, reporting only what a reader needs. Pasted output looks unfinished and signals that the numbers were not thought about.
The software has no view on whether your analysis makes sense. It will run:
None of these produce an error message. The output is indistinguishable from a valid result, which is why the analytic judgement has to come from you rather than from the absence of a warning.
Set the measurement level deliberately for every variable rather than accepting whatever the import defaulted to, declare your missing value codes explicitly, and add value labels so output is interpretable later. Then keep the raw file untouched and do all recoding on a working copy. Undeclared missing codes are the most damaging setup error: a “prefer not to say” coded as 99 and left undeclared is treated as the number ninety-nine, and every statistic involving that variable is wrong while the output looks entirely normal.
Use the menus if you prefer, but paste the syntax rather than running directly — every dialog offers this, and it costs nothing. Menu clicks leave no record, so six months later you cannot reconstruct whether cases were excluded before or after a scale was computed. Syntax gives you a reproducible record for your methodology chapter, lets you re-run an analysis if it is questioned, and means correcting an early error takes seconds rather than repeating hours of clicking.
Missing data. Most procedures exclude cases missing on any variable used in that particular analysis, so a model with several variables can run on noticeably fewer cases than you recruited. Check the sample size reported for each analysis, use that figure in your thesis rather than the recruitment number, and report how much data was missing and how you handled it.
That the p-value is smaller than the display precision, not that it is zero. Report it as less than .001 rather than as .000, which would claim an impossible probability. Where your reporting style asks for exact values elsewhere, follow it — this is the one case where a threshold statement is the correct form.
Better not to. Rebuild them in your discipline’s reporting format, including only what a reader needs — the test statistic, degrees of freedom, exact p-value, effect size and confidence interval where relevant. Pasted output includes tables that answer nothing, looks unfinished, and suggests the numbers were reproduced rather than interpreted.
No, and this is the single most important thing to understand about it. It will run a test whose assumptions your data violate, treat repeated measures as independent groups, or compute a mean of a nominal variable, and the output will look exactly like a valid result. There is no warning, so the judgement about whether an analysis suits your design has to come from you.
Most SPSS problems are set up before any test is run, and they are invisible in the output. Describe your data and what you are trying to establish, and a PhD in your field will go through it with you.
Leave an email or a WhatsApp number — whichever you prefer — and tick how we should reach you. We reach out within 30 minutes.
