RESEARCH SUPPORT
SPSS Assumption Checking and Diagnostics
Judged From Design and Residuals
Assumptions judged from the design of your study, the residual plots and a check of how far the conclusions move without them, not from one significance test, so that the model you report is one that a reviewer can accept and a committee can follow.
No Commitment Required
Strictly Confidential
In-House PhD-Level Statisticians
Fixed Project Fee
Same-Day Response
What Does SPSS Assumption Checking Involve?
SPSS assumption checking asks whether a fitted model is a defensible description of the data, and the answer comes from the design, the residuals and the sensitivity of the conclusions, not from a single test. Independence is a property of how the data were collected. Linearity, residual spread and the residual distribution are examined with plots, influential cases through saved Cook’s distance and leverage values, and collinearity through tolerance. SPSS model diagnostics inform a judgment: a departure matters when it changes the estimates or intervals, not merely when a test flags it. For OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, reading these checks is the work of in-house PhD-level statisticians. The choice of final model is agreed with the researcher.
WHEN TO ENGAGE
Triggers for an SPSS Diagnostic Review
A diagnostic review is usually requested when the validity of a model is in doubt, either because a reader raised a concern or because the plots look uncomfortable.
Plots Look Wrong A residual plot fans out, curves or shows a cluster of outliers, and it is unclear whether the pattern is serious enough to change the model or can be left alone.
Test Flags a Violation A Shapiro-Wilk or Levene result is significant in a very large sample, or not significant in a very small one, and neither result settles the question by itself.
Influential Cases Found One or two participants have a Cook's distance far above the rest, and the model must be judged with and without them, with the reason for any exclusion recorded.
Reviewer Asks for Evidence A reviewer or committee asks for evidence that the assumptions hold, and a statement that they were checked is not accepted as that evidence.
CHECKS AND JUDGMENTS
Diagnostics Matched to the Model Being Fitted
Which checks matter depends on the model and the design, so the checks follow them instead of a fixed checklist.
Independence From Design
Independence is judged from how participants were sampled and measured, and clustered or repeated data are modeled as such within Quantitative Data Analysis.
Linearity and Spread
SPSS residual analysis plots standardized residuals against predicted values to show curvature and changing spread, which are judged by eye and by what a remedy would change in the estimates.
Residual Distribution
A histogram and a normal probability plot of the residuals show skew and heavy tails, which affect inference mainly in small samples and for prediction intervals.
Influence and Leverage
Saved Cook's distance, leverage and DfBeta values show which cases move the coefficients, and the model is refitted without them, the same practice described under Multiple Linear Regression.
Overlap Among Predictors
Tolerance and variance inflation factors show where predictors overlap, and the response depends on why: product and polynomial terms overlap by construction, while related measures call for a substantive choice.
Reporting the Checks
What was checked, what was found and what was done is reported, and a model already fitted can be re-examined through Independent Statistical Review.
FALSE REASSURANCE
Misjudgments That Make a Model Look Safer Than It Is
Several habits give a false sense of security, and each of them can be replaced by a check that is both easier to defend and more informative.
One Significant Test, One Verdict A single p-value from Levene's test or a Shapiro-Wilk statistic is treated as the verdict, though in large samples trivial departures are flagged and in small samples serious ones are missed.
Uneven Spread Noted, Then Ignored A funnel in the residual plot is mentioned, yet the intervals are reported from the usual standard errors, although robust standard errors or a model for the spread would match the data better.
Outlier Removed to Improve Fit A case is dropped because it worsens the fit and not for any substantive reason, which makes the remaining data look cleaner than the process that produced them.
Log Scale Reported as Raw An outcome is log-transformed to tidy the plots, and the coefficients are then reported as if they were still on the original scale.
Durbin-Watson on Unordered Data A Durbin-Watson statistic is reported for a cross-sectional file whose row order carries no meaning, so the number says nothing about the independence that matters.
Checks Run, Results Not Shown Diagnostics are run and the model changed in response, but the paper reports only the final model, so a reader cannot see how far the result depended on those choices.
DELIVERABLES
The Diagnostic Evidence Set for Your Model
Every figure is paired with a plain statement of what it shows, which is the form a committee member or reviewer can use directly.
Residual Plot Panel Plots of standardized residuals against predicted values and against each predictor, each with a short reading of what it shows.
Influence Table The cases with the largest Cook's distance, leverage or DfBeta values, with the estimates refitted without them and the reason for each decision.
Collinearity Summary Tolerance and variance inflation values for every predictor, with a note on which overlaps are structural and which are substantive.
With-and-Without Comparison Key estimates and intervals from the original model beside those from the alternative specification, so that a reader sees how far the results move.
Commands for Every Check The SPSS commands that saved the residuals and influence measures and produced each plot, so the checks can be repeated on the final model.
Assumptions Note A short note on the assumptions, recording what was checked, what was found and what the final model reflects, for you to check and incorporate.
ILLUSTRATIVE EXAMPLE
A Skewed Outcome and Two Influential Fields
An illustration of how diagnostics informed the scale of a model and showed how far its conclusions depended on two cases.
ILLUSTRATIVE EXAMPLE · SOIL ORGANIC CARBON SURVEY RESEARCH · UNITED STATES
A soil science researcher modeled organic carbon concentration across farm fields from clay content, annual rainfall and tillage practice, and had to show that the linear model could be defended.
The first residual plot fanned out, with larger spread at higher predicted values, and the outcome was right-skewed. Two fields with unusually high clay content had a Cook’s distance far above the rest. A log transformation stabilized the spread, and the coefficients were then reported as percentage differences instead of on the original scale. The model was refitted without the two fields to show how far the estimates moved, and the fields were kept because they were genuine observations. Neighboring fields were likely to be alike, so independence was examined against the sampling design and not assumed, and the findings were described as associations.
Outcome: The researcher received the residual plots with readings, the influence table, the with-and-without comparison with 95% confidence intervals, the SPSS commands and an assumptions note.
This example does not describe a real client or project. We never share details of actual client work, in line with the confidentiality we promise every researcher. This illustrative example is included to show the type, scope, and standard of work involved in this kind of analysis.
SOFTWARE
SPSS Output and Its Counterparts Elsewhere
Diagnostics are not specific to one program, and a second tool is sometimes used to confirm a finding or to add a check beyond what the SPSS menus offer.
HOW WE ASSESS YOUR PROJECT
We Review Your Actual Research
Your assessment is based on a review of your actual research materials and analytical requirements, not a generic project category.
You Submit Your Project Details
Share whatever is available at the current stage, such as the research aim, protocol or proposal, dataset, variable list, manuscript or statistical requirements.
We Review Your Research and Define the Scope
Your statistically trained Research Coordinator reviews the materials to understand the study design, research questions, data structure, analytical requirements and expected deliverables. If clarification is needed, additional documents or information can be requested before the assessment is finalized.
Your Analysis Is Completed and Delivered
Once the scope and fixed project fee are agreed, the in-house statistician completes the work and the Research Coordinator remains the communication point through delivery, with the analysis, deliverables and timeline as agreed.
RESEARCH RESPONSIBILITY
Diagnostic Findings and Your Modeling Decisions
The checks and what they show are reported in full, and the decision about which model the study finally rests on is taken with the researcher.
Agreed Statistical Support OnlineSPSS provides the agreed methodological, analytical and statistical reporting support, defined in a scope that is confirmed before any work begins.
No Intellectual Contribution OnlineSPSS does not claim any intellectual contribution to your analysis, paper or thesis. Authorship remains the researcher's.
Your Research Decisions The research question, the substantive interpretation of the fitted model and the conclusions stay with you, as does overall responsibility for the study.
Confidential Handling Research data and project details are handled confidentially, project information is not published, and an NDA is available where required.
FREQUENTLY ASKED QUESTIONS
Logistic Models, Altered Models and Saved Residuals
A logistic model, a model altered after its residuals were inspected and the choice of saved residual each change what a diagnostic review has to explain.
What does assumption checking in SPSS involve beyond a normality plot?
SPSS assumption checking at OnlineSPSS covers independence from the design, linearity, the spread and distribution of the residuals, influential cases and collinearity, each judged by its consequence for the estimates. The checks follow the model being fitted, and a model that is not linear has its own set of diagnostics.
Which checks does a logistic model need instead of residual normality?
A logistic model has no assumption of normal residuals or constant variance. Its checks are the linearity of the logit for continuous predictors, influential cases, separation or too few events for the number of predictors, and the calibration of the predicted probabilities. Independence is again settled by the design and not by a statistic.
How is a diagnostic problem reported when the model is changed because of it?
The original model, the problem it showed and the changed model are all reported, with the estimates from both, since a model altered after the residuals were inspected is a data-dependent choice. A reader can then judge how much the conclusion rests on the change. Where the two sets of estimates agree, that agreement is itself useful evidence.
Which saved residual should be used to flag unusual cases in SPSS?
The Save option in the Linear Regression dialog offers several residuals. Standardized residuals divide by the overall error, whereas studentized deleted residuals come from a fit without the case and allow for its leverage, which suits flagging outliers. Cook’s distance shows whether a flagged case moves the coefficients.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the dataset, the model output and the residual plots you already have, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Test a Model Against
Its Own Assumptions
Residual plots and influence tables shown beside the estimates they affect let a reader judge the model directly, instead of taking on trust that the assumptions were checked.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014