OpenEpi Logo

Choosing

Choosing a Statistical Method, with Automatic Internet Searching

Taken from ‘Outline for the Proposed JavaScript Open Source Version of StatCalc’

By Hemant Kulkarni, San Antonio               November 18, 2002 Searching revised, June 2010

The following tables outline the suggested matrix. Please remember that the suggested options are the ones that CAN be chosen in a given situation. These options do NOT HAVE to be chosen.

To do a search, if you have an Internet connection, click on an item within a table cell. In some browsers, a choice of search will appear as a funny little icon; in others, the search may happen automatically, and may include ALL the items in the cell. After you see the results, you may want to add a word like ´method,´ ´calculator,´  ´equation,´  or ´tutorial´ to narrow the search. To return from the search, use the browser's BACK button or close the browser window.

  1. DESCRIPTIVE STUDIES

Situation

Epidemiologic results

Statistical methods

Time description

Scatter plots

Scatter plots with trend lines

Scatter plots with spline smoothing

Seasonality index

Place description

Spot maps

Clustering methods like k-means

Person description

Bar charts

Pie charts

Histograms

Box plots

Box and whisker plots

Stem and leaf diagrams

Hierarchical trees

Scatter plots

Mean

Proportion

Standard deviation

Standard errors

Median

Percentiles

Mode

  1. CROSS-SECTIONAL STUDIES

Situation

Epidemiologic results

Statistical methods

Group comparison

  • Risk reduction with 95% confidence intervals
  • Odds ratios with 95% confidence intervals
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • Cramer’s V
  • chi square test for linear trend
  • Student’s T test (unpaired)
  • Mann-Whitney U test
  • Mantel-Haenszel chi square test
  • Unconditional multiple logistic regression
  • Generalized linear models
  • Bonferroni’s corrections
  1. CASE-CONTROL STUDIES

Situation

Epidemiologic results

Statistical methods

Unmatched case-control study

  • Single exposure variable expressed as Yes/No
  • One set of controls

Odds ratio with 95% confidence interval

  • Cornfield’s method
  • Woolf’s method
  • Exact method
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient

Unmatched case-control study

  • Single exposure variable expressed as multiple, ordered categories
  • One set of controls

Odds ratio for each category compared to the reference category with 95% confidence interval

  • Cornfield’s method
  • Woolf’s method
  • Exact method
  • chi square test of association
  • Fisher’s exact test
  • Cramer’s V
  • chi square test for linear trend

Unmatched case-control study

  • Single continuous variable
  • One set of controls

-

  • Student’s T test (unpaired)
  • Mann-Whitney U test

Unmatched case-control study

  • Multiple exposure variables expressed as Yes/No
  • One set of controls

Odds ratio with 95% confidence interval for each variable separately

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Adjusted ORs for combinations of variables

Stratified and summary OR with 95% confidence interval

  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • Unconditional multiple logistic regression
  • Mantel-Haenszel chi square test

Unmatched case-control study

  • Multiple exposure variables expressed as two or more, ordered categories
  • One set of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Adjusted ORs for combinations of variable categories

  • chi square test of association
  • Fisher’s exact test
  • Cramer’s V
  • chi square test for linear trend
  • Unconditional multiple logistic regression

Unmatched case-control study

  • Multiple continuous variables
  • One set of controls

-

  • Student’s T test (unpaired)
  • Mann-Whitney U test
  • Unconditional multiple logistic regression

Unmatched case-control study

  • Combinations of variable types
  • One set of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Adjusted ORs for combinations of variable categories

Stratified and summary OR with 95% confidence interval

  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • Cramer’s V
  • chi square test for linear trend
  • Student’s T test (unpaired)
  • Mann-Whitney U test
  • Mantel-Haenszel chi square test
  • Unconditional multiple logistic regression
  • Generalized linear models

Unmatched case-control study

  • Single exposure variable expressed as Yes/No
  • Multiple sets of controls

Odds ratio with 95% confidence interval

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Stratified and summary OR with 95% confidence interval

  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • Mantel-Haenszel chi square test

Unmatched case-control study

  • Single exposure variable expressed as multiple, ordered categories
  • Multiple sets of controls

Odds ratio for each category compared to the reference category with 95% confidence interval

  • Cornfield’s method
  • Woolf’s method
  • Exact method

For each set of controls separately

  • chi square test of association
  • Fisher’s exact test
  • Cramer’s V
  • chi square test for linear trend
  • Bonferroni’s correction for multiple comparisons

Unmatched case-control study

  • Single continuous variable
  • Multiple sets of controls

-

  • Analysis of variance with Tukey’s pairwise comparison tests
  • Kruskal-Wallis test with Mann-Whitney test for pairwise comparisons after Bonferroni correction

Unmatched case-control study

  • Multiple exposure variables expressed as Yes/No
  • Multiple sets of controls

Odds ratio with 95% confidence interval for each variable separately

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Adjusted ORs for combinations of variables

Stratified and summary OR with 95% confidence interval

  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • Mantel-Haenszel test
  • Unconditional multiple logistic regression

Unmatched case-control study

  • Multiple exposure variables expressed as two or more, ordered categories
  • Multiple sets of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Adjusted ORs for combinations of variable categories

  • chi square test of heterogeneity
  • Fisher’s exact test
  • Cramer’s V
  • chi square test for linear trend
  • Unconditional multiple logistic regression

Unmatched case-control study

  • Multiple continuous variables
  • Multiple sets of controls

-

  • Analysis of variance with Tukey’s pairwise comparison tests
  • Kruskal-Wallis test with Mann-Whitney test for pairwise comparisons after Bonferroni correction
  • Polytomous multiple logistic regression

Unmatched case-control study

  • Combinations of variable types
  • Multiple sets of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • Cornfield’s method
  • Woolf’s method
  • Exact method

Adjusted ORs for combinations of variable categories

Stratified and summary ORs

  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • Cramer’s V
  • chi square test for linear trend
  • Student’s T test (unpaired)
  • Mann-Whitney U test
  • Unconditional multiple logistic regression
  • Analysis of variance and Tukey’s pairwise comparisons
  • Kruskal-Wallis test
  • Generalized linear models

Matched case-control study

  • Single exposure variable expressed as Yes/No
  • One set of controls

Odds ratio with 95% confidence interval

·       McNemar’s method

  • McNemar’s chi square test of association

Matched case-control study

  • Single exposure variable expressed as multiple, ordered categories
  • One set of controls

Odds ratio for each category compared to the reference category with 95% confidence interval by collapsing into 2X2 tables

  • McNemar’s method
  • McNemar’s chi square test of association
  • Breslow-Day test

Matched case-control study

  • Single continuous variable
  • One set of controls

-

  • Student’s T test (paired)
  • Wilcoxon signed-rank test

Matched case-control study

  • Multiple exposure variables expressed as Yes/No
  • One set of controls

Odds ratio with 95% confidence interval for each variable separately

  • McNemar’s Method

Adjusted ORs for combinations of variables

  • McNemar’s chi square test of association
  • Conditional multiple logistic regression

Matched case-control study

  • Multiple exposure variables expressed as two or more, ordered categories
  • One set of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • McNemar’s method

Adjusted ORs for combinations of variable categories

  • McNemar’s chi square test of association
  • Breslow-Day test
  • Conditional multiple logistic regression

Matched case-control study

  • Multiple continuous variables
  • One set of controls

-

  • Student’s T test (paired)
  • Wilcoxon signed- rank test
  • Conditional multiple logistic regression

Matched case-control study

  • Combinations of variable types
  • One set of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • McNemar’s method

Adjusted ORs for combinations of variable categories

  • McNemar’s chi square test of association
  • Breslow-Day test
  • Student’s T test (paired)
  • Wilcoxon signed-rank test
  • Conditional multiple logistic regression
  • Generalized linear models

Matched case-control study

  • Single exposure variable expressed as Yes/No
  • Multiple sets of controls

Odds ratio with 95% confidence interval

  • McNemar’s method

Stratified and summary OR with 95% confidence interval

  • McNemar’s chi square test of association
  • Mantel-Haenszel chi square test

Matched case-control study

  • Single exposure variable expressed as multiple, ordered categories
  • Multiple sets of controls

Odds ratio for each category compared to the reference category with 95% confidence interval by collapsing into 2X2 tables

  • McNemar’s method

For each set of controls separately

  • McNemar’s chi square test of association
  • Breslow-Day test
  • Bonferroni’s correction for multiple comparisons

Matched case-control study

  • Single continuous variable
  • Multiple set of controls

-

  • Friedman’s Analysis of variance
  • Kruskal-Wallis test with Mann-Whitney test for pairwise comparisons after Bonferroni correction

Matched case-control study

  • Multiple exposure variables expressed as Yes/No
  • Multiple sets of controls

Odds ratio with 95% confidence interval for each variable separately

  • McNemar’s method

Adjusted ORs for combinations of variables

Stratified and summary OR with 95% confidence interval

  • McNemar’s chi square test of association
  • Mantel-Haenszel test
  • Conditional multiple logistic regression

Matched case-control study

  • Multiple exposure variables expressed as two or more, ordered categories
  • Multiple sets of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • McNemar’s method

Adjusted ORs for combinations of variable categories

  • chi square test of heterogeneity
  • Breslow-Day test
  • Conditional multiple logistic regression

Matched case-control study

  • Multiple continuous variables
  • Multiple sets of controls

-

  • Friedman’s Analysis of variance
  • Kruskal-Wallis test with Mann-Whitney test for pairwise comparisons after Bonferroni correction
  • Polytomous multiple logistic regression

Matched case-control study

  • Combinations of variable types
  • Multiple sets of controls

Odds ratio for each category compared to the reference category with 95% confidence interval for each variable separately

  • McNemar’s method

Adjusted ORs for combinations of variable categories

Stratified and summary ORs

  • McNemar’s chi square test of association
  • Breslow-Day test
  • Student’s T test (paired)
  • Wilcoxon’s signed-rank test
  • Conditional multiple logistic regression
  • Friedman’s Analysis of variance
  • Kruskal-Wallis test
  • Mantel-Haenszel test
  • Generalized linear models

4. COHORT STUDIES

Situation

Epidemiologic results

Statistical methods

Outcome is number of events in the exposed and unexposed groups where the event can occur only once in a single individual

  • Relative risk with 95% confidence interval
  • Excess fraction (also called as risk reduction)
  • Etiologic fraction (also called attributable risk)
  • Population attributable risk
  • Population attributable risk proportion
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • chi square test of linear trend for multiple ordered categorical variables
  • Unconditional multiple logistic regression
  • Generalized linear models

Outcome is number of events in the exposed and unexposed groups where the event can occur multiple times in a single individual

  • Relative hazards and their 95% confidence interval
  • Kaplan-Meier survival curves
  • Log-rank test OR Wilcoxon test based on the violation of proportional hazards assumption
  • Cox proportional hazards model
  • Poisson’s multiple regression

Outcome is time to events in the exposed and unexposed groups

  • Relative hazards and their 95% confidence intervals
  • Kaplan-Meier survival curves
  • Log-rank test OR Wilcoxon test based on the violation of proportional hazards assumption
  • Cox proportional hazards model

Outcome is a continuous variable for the exposed and unexposed groups

  • Relative risks and their 95% confidence intervals using clinically meaningful cut-offs
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • chi square test of linear trend for multiple ordered categorical variables
  • Unconditional multiple logistic regression
  • Generalized linear models
  • Time series analyses

5. CLINICAL AND PREVENTIVE TRIALS

Situation

Epidemiologic results

Statistical methods

Outcome is number of events in the exposed and unexposed groups where the event can occur only once in a single individual

  • Relative risk with 95% confidence interval
  • Excess fraction (also called as risk reduction)
  • Etiologic fraction (also called attributable risk)
  • Number needed to treat (NNT)
  • Vaccine efficacy with its 95% confidence interval
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • chi square test of linear trend for multiple ordered categorical variables
  • Unconditional multiple logistic regression
  • Generalized linear models

Outcome is number of events in the exposed and unexposed groups where the event can occur multiple times in a single individual

  • Relative hazards and their 95% confidence interval
  • Vaccine efficacy with 95% confidence interval
  • Kaplan-Meier survival curves
  • Log-rank test OR Wilcoxon test based on the violation of proportional hazards assumption
  • Cox proportional hazards model
  • Poisson’s multiple regression
  • Sequential trial with interim analyses using the alpha-spending function
  • Generalized linear models

Outcome is time to events in the exposed and unexposed groups

  • Relative hazards and their 95% confidence intervals
  • Vaccine efficacy and its 95% confidence interval
  • Kaplan-Meier survival curves
  • Log-rank test OR Wilcoxon test based on the violation of proportional hazards assumption
  • Cox proportional hazards model

Outcome is a continuous variable for the exposed and unexposed groups

  • Relative risks and their 95% confidence intervals using clinically meaningful cut-offs
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • chi square test of linear trend for multiple ordered categorical variables
  • Unconditional multiple logistic regression
  • Generalized linear models
  • Time series analyses
  1. SCREENING TEST PERFORMANCE EVALUATION

Situation

Epidemiologic results

Statistical methods

Dichotomous (Positive/Negative) result compared with a reference standard

  • Sensitivity
  • Specificity
  • Positive predictivity
  • Negative predictivity
  • Accuracy
  • Likelihood ratio of a positive test
  • Likelihood ratio of a negative test
  • Cohen’s kappa
  • Entropy
  • Bias index
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient

Multiple categorical test outcome compared with a reference standard

  • Likelihood ratios for each categorical test outcome
  • Receiver-operating characteristic curve
  • chi square test of association
  • Fisher’s exact test
  • Φ coefficient
  • chi square test of linear trend
  • Area under the ROC curve and its 95% confidence interval by Wilcoxon’s method

Continuous test outcome compared with a reference standard

  • Likelihood ratios for pre-defined categories of the outcome
  • Receiver operating characteristic curve
  • Area under the ROC curve and its 95% confidence interval
  • Optimum operating point (OOP) choosing
  • Cost-of-error comparisons