Home › Statistics Homework Help › SPSS Help › Finding Outliers in SPSS
Find it, check it, report it

How do I find and check for outliers in SPSS?

You find outliers in SPSS three ways: read a boxplot from Analyze > Descriptive Statistics > Explore, where SPSS marks a mild outlier with a circle and an extreme value with a star; save standardised z-scores from Descriptives and flag any absolute value above 3.29; and for several variables at once, save Mahalanobis distance from a linear regression and compare it against a chi-square critical value. Finding the point is step one. This page also shows how to check whether it is real and what to do with it.

✓Boxplot, z-score and Mahalanobis ✓How to handle and report them ✓Human statisticians, no AI

On this page

Boxplots in Explore1.5x and 3x IQRStandardised z-scores |z| > 3.29Mahalanobis distanceChi-square cut-off Handling outliersReportingWorked example
Since 2014statistics specialists
3 methodsunivariate and multivariate
★★★★★rated by students
3 hrsfastest turnaround

An outlier is a value that sits far away from the rest of your data. It matters because a single extreme point can pull a mean, inflate a standard deviation, weaken a correlation or tilt a regression line, so most marking schemes expect you to screen for outliers before you run the main test. This guide is part of our SPSS homework help. It gives the exact menu path for each screening method, points at the number that decides whether a case is an outlier, and shows how to check the point and report your decision.

One idea to keep straight from the start. Finding an outlier is not the same as deleting one. SPSS flags candidate points for you, but the flag is a prompt to investigate, not an instruction to erase the row. The second half of this page is about that judgement, because that is where the marks and the honest analysis live.

Which method do I need?

Match how many variables you are screening to the right SPSS method, then jump to its steps.

Your situationWhat you are checkingMethod
One continuous variable, want a visual flagValues outside 1.5 and 3 box-lengthsBoxplot in Explore
One continuous variable, want a numberStandardised distance from the meanStandardised z-scores
Several continuous variables togetherUnusual combination across variablesMahalanobis distance

A point can be ordinary on each variable alone yet unusual in combination, which is why multivariate work needs Mahalanobis distance, not just a column of z-scores.

Before you screen: set the measure and know the two windows

Outlier screening applies to continuous variables, so check the Measure column in Variable View first. SPSS tags every numeric variable as Scale by default, and that is what you want for a genuine continuous measure such as height, score or reaction time. A variable that is really a category coded as a number, for example one for male and two for female, is not a candidate for outlier screening at all, so set it to Nominal and leave it out of this process.

Everything below happens through the Analyze menu on the data you can see in Data View. The boxplot and z-score methods look at one variable at a time, and Mahalanobis distance looks at a set of variables together. None of them change your data unless you choose to act on what they find.

How to find outliers in SPSS, three methods

Menu path, the rule SPSS uses, the exact value to read, and the recreated output. Example values follow standard teaching datasets.

1. Boxplots via Analyze > Descriptive Statistics > Explore

The fastest visual screen for one continuous variable. SPSS labels the suspect cases for you with their case numbers.

Analyze > Descriptive Statistics > Explore. Move your continuous variable into the Dependent List. Click Plots and make sure Boxplots is set to Factor levels together. Click Statistics and tick Outliers, which prints the five highest and five lowest cases. Continue, then OK.

The rule SPSS uses. SPSS builds the box from the first and third quartiles, so the box length is the interquartile range, the IQR. It then draws whiskers to the most extreme values that are still within 1.5 box-lengths of the box. Beyond that, a value between 1.5 and 3 box-lengths from the edge of the box is a mild outlier, drawn as an open circle with its case number beside it. A value more than 3 box-lengths out is an extreme value, drawn as a star or asterisk with its case number. The number next to the symbol is the row in Data View, not the value, so you can find and inspect that case.

What to read. The boxplot itself for the circles and stars, and the Extreme Values table, which lists the five highest and five lowest cases with their case numbers and actual values so you can confirm what the plot flagged.

Extreme Values (reaction time, ms)
Case NumberValue
Highest127842
214561
38548
Lowest119301
23318

Here case 27 is drawn as a star on the boxplot, well beyond 3 box-lengths above the box, so it is an extreme value to investigate. Case 14 sits as a circle, a milder outlier.

Common mistakes

  • Reading the case number as the data value. The label by a circle or star is the row number, so you still open that case to see the value.
  • Treating every circle as a reason to delete a row. A circle is a mild outlier and often a real observation.
  • Running Explore split by a factor you did not mean to, so the plot shows within-group outliers when you wanted the whole sample.

2. Standardised z-scores via Analyze > Descriptive Statistics > Descriptives

Turns each value into its distance from the mean in standard deviations, giving a single number you can threshold.

Analyze > Descriptive Statistics > Descriptives. Move your continuous variable into Variable(s). Tick Save standardized values as variables at the bottom left, then OK. SPSS adds a new column named with a Z prefix, for example Zreaction, holding the z-score for every case.

The rule. A z-score says how many standard deviations a value sits from the mean, so the sign shows direction and the size shows distance. The widely taught cut-off is an absolute z above 3.29, because a value that far out lies beyond the middle 99.9 percent of a normal distribution, which corresponds to p below .001. Some courses use a rounder cut-off of 3. Sort the new Z column, or use Data > Select Cases with the condition ABS(Zreaction) > 3.29, and inspect every case that passes it.

What to read. The new Z column in Data View. Any case whose absolute z-score exceeds your threshold is a potential univariate outlier.

Cases flagged by Zreaction (threshold 3.29)
Case NumberReaction (ms)ZreactionFlag
278423.71Outlier, p < .001
145612.18Within range
19301-1.94Within range

Common mistakes

  • Comparing the raw z-score to 3.29 without taking the absolute value, so a large negative outlier is missed.
  • Using z-scores on a badly skewed variable, where many honest values look extreme because the mean and standard deviation are themselves pulled by the skew.
  • Forgetting that z-scores standardise one variable at a time, so they do not catch a case that is unusual only in combination.

3. Mahalanobis distance for multivariate outliers

Finds a case that is unusual across several variables at once, even when it looks ordinary on each variable alone. Needed before regression or factor analysis.

Analyze > Regression > Linear. Put any variable, even the case ID, in Dependent, and move the set of continuous variables you are screening into Independent(s). Click Save, tick Mahalanobis under Distances, Continue, then OK. SPSS writes a new column named MAH_1.

The rule. Mahalanobis distance measures how far a case sits from the centre of the cloud of points, adjusting for how the variables correlate. You compare each MAH_1 value against the chi-square critical value at a strict alpha, usually p below .001, with degrees of freedom equal to the number of predictors you entered. A case whose Mahalanobis distance is larger than that critical value is a multivariate outlier. For example, with three predictors the chi-square critical value at p below .001 is about 16.27, so any MAH_1 above 16.27 is flagged.

What to read. The MAH_1 column, sorted in descending order, against your chi-square critical value. You can also use Data > Select Cases with the condition MAH_1 > 16.27 to isolate them.

Mahalanobis distance, 3 predictors, critical value 16.27 (p < .001)
Case NumberMAH_1Decision
2718.94Multivariate outlier
4112.06Retain
97.55Retain

Common mistakes

  • Using the wrong degrees of freedom. The df is the number of predictors entered, not the sample size.
  • Comparing MAH_1 to 3.29. That is the z-score rule; Mahalanobis distance uses a chi-square critical value.
  • Screening multivariate outliers with separate z-scores, which misses a case that is only unusual in combination.

Not sure which method fits your design, or which critical value to use for your number of predictors? Our statisticians will screen your file and interpret it. Get a quote →

What to do with an outlier once you find it

A flag from any of the three methods is the start of a decision, not the end of one. Work through it in order.

  1. Investigate for a data-entry or measurement error first. Open the flagged case and check the value against the source. A reaction time of 8420 where the range is 300 to 900 is almost certainly a misplaced digit. If it is an error you can correct, fix it. If it is an error you cannot recover, remove that value and say so in your write-up.
  2. If the value is genuine, keep it and report it. A real extreme observation is data. The honest default is to run the analysis both with and without the point and report whether the conclusion changes. If it does not change, the outlier is not driving your result and you keep it.
  3. Consider a transformation. When a variable is skewed and the outliers are the long tail of that skew, a log or square-root transformation in Transform > Compute Variable can pull the tail in and make the extreme values ordinary, without deleting anything.
  4. Consider a resistant method. If the outlier is real and you do not want it to dominate, use a method that resists extreme values, such as reporting the median and interquartile range instead of the mean, using a non-parametric test, or fitting a resistant regression.

How to report it. State the screening method and threshold you used, how many cases were flagged, what you decided for each and why, and whether the main result held with and without them. One or two honest sentences in the method section is what marking schemes look for. Deleting points quietly is what they penalise.

Report it (method section)Data were screened for outliers using boxplots and standardised z-scores. One case exceeded an absolute z of 3.29 (z = 3.71). On inspection the value was a plausible response rather than an entry error, so it was retained. The independent-samples t-test was run with and without this case and the conclusion was unchanged, so results are reported for the full sample.

A short worked example

Suppose you have reaction times for 30 participants and you plan an independent-samples t-test. You run Analyze > Descriptive Statistics > Explore on the reaction-time variable and the boxplot shows one star at the top, labelled case 27, and one circle, labelled case 14. To put numbers on it, you run Descriptives with Save standardized values as variables. The new Zreaction column shows case 27 at z = 3.71, above the 3.29 cut-off, and case 14 at z = 2.18, inside it. So the boxplot and the z-scores agree on one univariate outlier, case 27.

You open case 27. The value is 842 ms, which is high but physically possible for that task, and the rest of that participant's data looks normal, so it is not an entry error. You retain it, then rerun the t-test with and without case 27. The p-value moves from .020 to .028, still significant, so the outlier is not driving the result. You keep the full sample and note the check in your method section. That is a complete, defensible outlier screen: find it, check it, decide, report.

Outliers versus influential points, and other traps

Procedures on this page were checked against the IBM SPSS Statistics documentation, the UCLA statistical computing SPSS guides and university statistics tutorials from Laerd Statistics and Kent State University.

Or hand the outlier screen to a statistician

When the flags disagree, the variable is skewed, or you are not sure whether to keep a point, a specialist screens it in SPSS on your file and explains the decision so you can defend it.

What one order includes: the right screening method for your design, run in SPSS on your data, the boxplots, z-scores or Mahalanobis distances, a plain-language call on each flagged case, and the result written in your required reporting style. A syntax file is included so the screen is reproducible.
1

Send your SPSS task

Upload your data file, the assignment brief and your deadline, then get a free quote from support.

2

Pay securely

Approve the price and a statistician who works in SPSS every day starts right away.

3

Download the analysis

Get the output, interpretation and write-up in your account, with free revisions if you need them.

Vetted and safe

Real statisticians, and your data protected

Your work goes to a specialist who passed a subject-specific statistics test and a background check, and who works in SPSS every day. What we deliver is a model answer and study aid for reference, and many students use it to learn the method and check their own analysis.

Everything stays private. Your data and details are confidential, we never contact your school, payments are secure, and the work is original and plagiarism-checked.

Get SPSS help you can trust →

Every order includes

Right method for the designOutput and interpretation Syntax file included100% human, no AI Data kept confidentialFree revisions
★★★★★
What students say
★★★★★

"Very good and friendly customer service. Most importantly give great results on time. Im very I chose homeworkdoer to do work for me. Excellent service."

Panos X.
Verified review
★★★★★

"Such great customer service and really helped out when I was in a jam for time and made awesome grades. Thank you so much."

Jamie M.
Verified review

Outliers in SPSS FAQ

In a boxplot from Analyze > Descriptive Statistics > Explore, SPSS marks a mild outlier with an open circle and the case number next to it, and an extreme value with a star or asterisk and its case number. A circle sits between 1.5 and 3 box-lengths from the edge of the box, and a star sits more than 3 box-lengths out, where a box-length is the interquartile range.

A common rule is that a standardised value with an absolute z above 3.29 is a potential outlier, because that point lies beyond the middle 99.9 percent of a normal distribution, which is p below .001. Some courses use the simpler cut-off of 3. Save z-scores with Analyze > Descriptive Statistics > Descriptives and the Save standardized values as variables tick box, then scan the new Z column.

Not by default. First investigate the value. If it is a data-entry or measurement error, correct it or remove it and say so. If it is a genuine observation, the usual practice is to keep it, run the analysis with and without it, and report both, or to use a transformation or a resistant method. Deleting a real value only because it is inconvenient is not acceptable and must be justified.

Use Mahalanobis distance. Go to Analyze > Regression > Linear, put any variable in Dependent and your set of continuous variables as Independents, then in Save tick Mahalanobis under Distances. SPSS writes a MAH_1 column. Compare each value against the chi-square critical value at your alpha, using degrees of freedom equal to the number of predictors. A case above that critical value is a multivariate outlier.

An outlier is a value far from the rest of the data on one or more variables. An influential point is a case that noticeably changes the model, for example the regression slope, when it is included. A point can be an outlier without being influential, and a case with high leverage can be influential without looking extreme on any single variable. In regression you check influence with Cook's distance, not with a z-score.

No. A large sample helps parametric tests tolerate mild non-normality through the central limit theorem, but it does not neutralise a genuine extreme value, which can still distort a mean, a standard deviation, a correlation or a regression slope. You check for outliers regardless of sample size.

Get your SPSS done right

Hand it to a statistician and get accurate screening and a clear interpretation before your deadline.

Get my free quote →

Related homework help

We use cookies to improve your experience. See our cookie policy.