My results don't make sense
Why Did My Regression Coefficient Change Sign?
Separate suppression, confounding, multicollinearity, and coding problems.
THE ANALYSIS CLINIC
Something went wrong
with your analysis.
Let's figure out why.
Statistical models fail. Results contradict expectations. Assumptions get violated. Committees ask unexpected questions. The Analysis Clinic explains what the symptoms mean, what usually causes them, and what to do next.
What are you dealing with?START WITH THE PROBLEM
What are you dealing with?
Choose the description that sounds most like where you are stuck. Each department leads to focused diagnostic Cases rather than a general methods archive.
My model won't run
Convergence warnings, identification failures, singular fits, separation, and models that will not estimate or behave as expected.
Convergence · identification · singular fits · separation See relevant CasesMy results don't make sense
Coefficients, tests, effects, intervals, and model comparisons that appear contradictory or are easy to misinterpret.
Coefficients · interactions · unexpected effects See relevant CasesSomething looks wrong with my data
Missingness, outliers, skewness, sparse cells, unequal groups, dependence, and other conditions that may affect the analysis.
Missingness · outliers · skewness · sparse cells See relevant CasesI may have chosen the wrong analysis
Choosing between plausible methods when the outcome, design, nesting, measurement, or inferential goal determines the defensible approach.
Mixed models vs. ANOVA · SEM vs. regression See relevant CasesMy committee or reviewer challenged my analysis
Questions about power, covariates, clustering, sensitivity analyses, effect sizes, and decisions a researcher must explain and defend.
Power · covariates · clustering · sensitivity See relevant CasesFIRST ON THE EXAM TABLE
Four diagnosed launch Cases.
Cases 001–004 are diagnosed and ready to read. Upcoming Cases become clickable only after their full diagnosis, reproducible example, and committee guidance pass review.
My model won't run
Why Won't My SEM Model Converge?
Understand what nonconvergence can signal before changing the estimator, deleting parameters, or simplifying the model.
My model won't run
Why Am I Getting a Singular Fit in My Mixed Model?
Diagnose what the estimated random-effects structure is telling you—and what it is not.
My model won't run
Perfect Separation in Logistic Regression
Recognize separation, locate its source, and compare defensible estimation and reporting options.
My results don't make sense
My Interaction Is Significant, but the Main Effects Aren't
Interpret conditional effects without treating nonsignificant main effects as a contradiction.
MORE THAN A WARNING FIX
Understand the decision.
Defend the reasoning.
The Clinic is designed to help dissertation researchers understand what a statistical problem means, diagnose its likely source, and choose a response that fits the research question and design. Removing a software warning is not the same as solving the analytical problem.
If an SEM problem requires review of your syntax and full output, the established SEM Rescue Clinic offers individualized diagnosis without asking you to send raw participant data.
THE CASE PIPELINE
Diagnoses in development.
Each Case belongs to one primary department. COMING SOON cards are intentionally non-clickable until the educational Case is complete.
I may have chosen the wrong analysis
My ICC Is Small. Do I Still Need Multilevel Modeling?
Connect clustering, the estimand, and uncertainty instead of using one cutoff.
My committee or reviewer challenged my analysis
My Sample Is Smaller Than Planned
Distinguish what can still be learned from claims the achieved sample cannot support.
Something looks wrong with my data
My Reliability Is Lower Than Expected
Check scoring, dimensionality, item behavior, and the limits of a single coefficient.
My model won't run
Why Is My SEM Model Fit Poor?
Investigate misfit without chasing indices or making unsupported modifications.
Something looks wrong with my data
Missing Data: Deletion, Multiple Imputation, or FIML?
Match the missing-data strategy to the analysis, assumptions, and available information.
Something looks wrong with my data
My Groups Are Extremely Unequal
Assess precision, assumptions, estimability, and what unequal groups change.
My committee or reviewer challenged my analysis
My Committee Wants More Covariates
Choose adjustments based on the research question and design, not a longer variable list.
BRING YOUR MODEL TO THE ANALYSIS CLINIC
Still stuck after
the first checks?
Some problems turn on details of the design, sample, model specification, or committee request. A consultation can help identify the next defensible step.
Book a free consultation