Interactive Inferential

Statistics Flowchart
Version 1.0
Last Updated: 8 January 2019
Directions
• View this in “Presentation Mode” in MS PowerPoint
• Click the “Begin” button below to start
• Click the desired measure of interest (e.g., Measure of Center)
• Follow the flowchart to arrive at the recommended inferential statistical
method
• Use the button to return to the previous slide
• Use the button to return to the starting page
• Questions? Contact the STAT COE at [email protected]

Begin
 Measure of
Center

Percentile Continuous Measure

(e.g., accuracy or performance
measures)

Measure of
Spread
What To Test?

Proportion Binary Measure

(e.g., Probability of Success, Phit,
Pmiss)

Relationship
between 2 or more
variables
Continuous or Binary Measure

Distributions
References
Measure of Center
STAT COE Best Practices: Popn variance 1-sample z-
(𝜎) known test/CI
Compare 1
Statistics Reference Series Part 1: Descriptive population mean
Statistics to threshold
Popn variance 1-sample t-
Interpreting Confidence Intervals (𝜎) unknown test/CI
Statistical Hypothesis Testing
2-sample t-
Understanding Analysis of Variance Equal variances test/CI (pooled
Populations are variance)
Normality
Assumption independent 2-sample t-
Compare 2 Unequal
Holds test/CI (unequal
population variances
variances)
means
Paired t-test
Populations are
paired (1-sample t-test
on differences)
Compare 3+
population ANOVA
means
Measure of
Center Popn dist is
Sign Test
continuous
Compare 1
population median
to threshold Popn dist is
Wilcoxon Signed
continuous and
Rank Test
symmetric

Wilcoxon Rank-
Populations are
Sum Test, Mann-
Normality Compare 2 independent Whitney Statistic
Assumption population
does NOT hold medians Wilcoxon Signed
Populations are
Rank Test/Sign
paired
Test (on diffs)
General
Kruskal-Wallis
treatment
Compare 3+ Error terms have Test
comparison

References
population
medians
same continuous
dist Test for ordered
Jonkheere-
Exit
treatment
Terpstra Test
effects
Percentile
STAT COE Best Practices:
Confidence Intervals for the Median and Other
Percentiles

See Measure
Median
of Center

Percentile CIs for

Percentile
Not Median
Bootstrap
CIs/Hyp Tests

References Exit
Measure of Spread
STAT COE Best Practices:
Statistics Reference Series Part 1: Descriptive
Statistics Compare 1
Chi-Square test
popn variance
for variance
to threshold*
Normality Compare 2 F-test for ratio
assumption popn of two
Holds variances* variances
Test equality of Levene’s test
≥2 popn Brown-
Variance/Std. variances Forsythe Test
Deviation
Compare 1 Bootstrap
popn variance confidence
Measure of to threshold intervals
Normality Try response
transformation to
Spread assumption apply normal-theory Medians are Ansari-Bradley
does NOT hold methods first
equal Test
Bootstrap Compare 2
Other (e.g., popn variances
confidence Medians are
Range, IQR) Jackknife Test
intervals NOT equal

References
*For CIs of the standard deviation (𝜎), simply take the square root of the confidence bounds for variance (𝜎 2 ) Exit
Proportion
STAT COE Best Practices: Adjusted (Agresti-Coull)
𝑛𝑝 ≥ 5, 𝑛ሺ1 − 1-sample proportion
Interpreting Confidence Intervals test/CI;
Statistical Hypothesis Testing
Score confidence
Using OC Curves to Balance Cost and Risk Compare 1
interval
population
proportion to
threshold Score confidence
Above interval;
assumptions fail Clopper-Pearson
Test/CI (exact method)

2-sample
Large sample proportion test
(on difference)
Independent
samples Adjusted 2-sample
Proportion proportion test (add 1 to
Small sample each cell in contingency
table);
Compare 2
population Fisher Exact Test
proportions
Compare marginal
proportions
McNemar Test
Dependent samples
(matched pairs) Compare stratified
contingency tables Cochran-Mantel-
Pearson’s chi- (have a confounding Haenszel Test
Compare >2 factor)
square test for
population
References proportions
homogeneity of
proportions
Exit
Relationship between 2 or more variables

Response is
continuous
(click to continue)
Relationship between
2 or more variables
Response is discrete
(click to continue)

References Exit
Relationship between 2 or more variables (Continuous Response)
STAT COE Best Practices:
Interpreting Confidence Intervals Simple linear regression;
Predictor is
Pearson’s correlation
Statistical Hypothesis Testing continuous
1 predictor coeff
Understanding Analysis of Variance
variable only
Model Building Process Part 1 Predictor is
Response ANOVA
Model Building Process Part 2 normality
categorical
Using Regression Trees assumption holds
2+ predictor Linear
variables regression

Linear regression on
Test for transformed response;
association GLM; Spearman’s rank
Relationship between
Predictor is correlation coeff
continuous response and
1+ predictor variables continuous
Test for Kendall correlation
1 predictor independence statistic
variable only

Predictor is Kruskal-Wallis
categorical Test
Normality
assumption does Linear regression on
NOT hold transformed response;
generalized linear model (GLM)
Linear regression
One categorical
Friedman
factor, one
2+ predictor Statistic
blocking factor
variables
Smoothers, splines,
Non-parametric
wavelets, Gaussian
regression
References process model Exit
Relationship between 2 or more variables (Discrete
Response)

Predictor is Logistic
continuous Regression

Pearson’s Chi-
1 variable only Large sample Square Test;
(all expected
frequencies >5) Likelihood-Ratio
Response is Predictor is Test of Statistic
binary categorical Independence
Small Sample
(some expected Fisher’s Exact Test
Logistic frequencies <5)
2+ variables
regression
Relationship
between
Nominal
discrete Response is
logistic
response and nominal
regression
1+ predictor
variables
Response is Ordinal logistic
ordinal regression

Response is a Poisson
count (Poisson) regression

References Exit
Distributions

Test if distribution Probability plots

follows particular Pearson’s Chi-Square
probability distribution Goodness of Fit Test

Compare equality of Kolmogorov-Smirnov

Distribution
two distributions Test

Shapiro-Wilk
Test for normality
Anderson-Darling

References Exit
References

• Agresti, Alan and Brent A. Coull. “Approximate is Better than ‘Exact’ for
Interval Estimation of Binomial Proportions.” The American Statistician, vol.
52, no. 2, 1998, pp. 119-126.
• Hines, W., Montgomery, D.C., Goldman, D.M., and Borror, C.M. Probability and
Statistics in Engineering. John Wiley & Sons, Inc., 2008.
• Agresti, Alan. Categorical Data Analysis. 3rd Ed. John Wiley & Sons, Inc., 2013.
• Hollander, M., Wolfe, D.A., Chicken, E. Nonparametric Statistical Methods. 3rd
Ed. John Wiley & Sons, Inc., 2014.

Exit