Applied Multivariate Stats Social Sciences Stevens
One book, placed in its field — what this book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-applied_multivariate_stats_social_sciences_stevens
Arrow keys or swipe to move · N for notes · F for full screen
Applied Multivariate Stats Social Sciences Stevens
One book, placed in its field — what this book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-applied_multivariate_stats_social_sciences_stevens
The problem
Soundness of inference erodes when it is left to instinct instead of method.
You were taught the moves piecemeal, never the whole model.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
The model
- 01Research Design Quality
- 02Subject-per-Variable Ratio
- 03Judicious Variable Selection
- 04Data Screening Rigor
- 05Statistical Power
- 06Model Assumption Tenability
- 07Capitalization on Chance
- 08Result Generalizability
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
Step 1 of 5 · Research Design Quality
Random assignment protects internal validity; representative sampling protects external validity — no single design maximizes both.
Misconception
Practitioners believe that sophisticated multivariate methods like ANCOVA or SEM can statistically 'control away' the weaknesses of a nonrandomized or observational design.
Reality
Statistical control adjusts only for the confounders you measured and modeled correctly; unmeasured or mismeasured confounds remain, so covariate adjustment in a quasi-experiment is a partial fix, not a substitute for random assignment.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
Step 2 of 5 · Subject-per-Variable Ratio
The 15:1 heuristic is a minimum for stable estimation, and stepwise methods require substantially more.
Misconception
Practitioners assume the 15:1 (or 10:1) rule is a fixed law that guarantees stable results once met.
Reality
The needed ratio rises with the technique's greediness and the number of parameters it estimates — stepwise regression and factor analysis demand far more cases per variable than a simple simultaneous regression, and a met ratio is a floor, not a certificate of stability.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
Step 3 of 5 · Judicious Variable Selection
Justify variables before collecting data, not after inspecting results.
Misconception
Practitioners believe that including more variables makes a model more thorough and reduces the risk of omitting something important.
Reality
Adding variables without justification increases capitalization on chance and multicollinearity while diluting power; a smaller theoretically grounded set typically predicts new samples better than a large kitchen-sink model.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
Step 4 of 5 · Data Screening Rigor
Multivariate outliers require Mahalanobis distance; univariate checks alone will miss them.
Misconception
Practitioners think data screening means little more than checking for out-of-range values and impossible codes.
Reality
Thorough screening also detects multivariate outliers invisible in any single variable, influential points that bend the fitted model, and patterns of missingness that bias results — cases that pass univariate checks can still be corrosive.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
Step 5 of 5 · Statistical Power
Power depends jointly on N, effect size, and alpha — fix the target effect size first.
Misconception
Practitioners interpret a non-significant p-value as evidence that the effect is absent or negligible.
Reality
Non-significance in a low-power study is uninformative — it fails to distinguish 'no effect' from 'real effect we couldn't detect,' which is why power must be estimated before, not after, the study.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
The cheat sheet
- Research Design QualityRandom assignment protects internal validity; representative sampling protects external validity — no single design maximizes both.
- Subject-per-Variable RatioThe 15:1 heuristic is a minimum for stable estimation, and stepwise methods require substantially more.
- Judicious Variable SelectionJustify variables before collecting data, not after inspecting results.
- Data Screening RigorMultivariate outliers require Mahalanobis distance; univariate checks alone will miss them.
- Statistical PowerPower depends jointly on N, effect size, and alpha — fix the target effect size first.
- Model Assumption TenabilityRobustness has limits, and covariance-homogeneity violations with unequal N are among the most dangerous.
- Capitalization on ChanceStepwise selection optimizes to sample noise and produces upwardly biased fit statistics.
- Result GeneralizabilityStatistical significance and generalizability are different questions with different remedies.
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data
Sources
- Applied Multivariate Stats Social Sciences Stevens
Applied Multivariate Stats Social Sciences Stevens · Mike West, Editor · Leading With Data