Epi 3: Selecting Statistical Tests Flashcards
List 4 different categories that answer question # 2 in selecting correct statistical test.
Q2 = What type of comparison/assessment is desired?
- Correlation - 2 groups
- Regression - prediction, 3+ groups
- Survival Comparison - time
- Group Comparison
2 vs. 3+ groups;
independent vs. paired/related
“prediction” indicates which category of statistical tests
Regression
3 regression tests = Logistic Regression, Multinominal Logistic Regression, Linear Regression
What does correlation statistical test compare?
Quantitative representation of STRENGTH + DIRECTION of a relationship between TWO variables
FIXED value range = [-1 - +1]
\+/- = direction value = magnitude of strength
How many Correlation statistical tests? Are any specific to particular data level?
Contingency Coefficient (nominal) Spearman Correlation (ordinal) Pearson Correlation (interval) Kappa Statistic
For Pearson Correlation test, what does a statistically significant value (p < 0.05) and a non-statistically significant value (p > 0.05) indicate?
p < 0.05: Linear correlation is present between 2 groups
p > 0.05: NO linear correlation, but may be a non-linear relationship present!!!
Under what circumstances would you run a Pearson Correlation as a “partial correlation”? What level of data would this involve?
Partial Correlation = correlation stat. test that controls for confounding variables
Only interval data = Pearson Correlation
Contingency Coefficient
Correlation stat. test
Nominal data
Spearman Correlation
Correlation statistical test
Ordinal data
Pearson Correlation
Correlation stat. test
Interval data
Kappa Statistic
A correlation; measures STRENGTH + DIRECTION of agreement between different evaluators;
Compares ability of different evaluators to classify/interpret same data into 1 of many groups.
Ex.- 2 different radiologists interpret/classify the same radiographic image
FIXED value range: [-1 - + 1]
+/- = good agreement/poor agreement
What does a regression measure? What else can be calculated from a regression?
Measure of relationship between multiple variables (IV) specific ability to predict outcome variable (DV)
Odds ratio for measure of association + control for confounding variables
How determine the data level of a Regression?
= data level of outcome variable (DV) = variable being predicted
What type of data comparison group can calculate Odds Ratio (OR)?
regression
Nominal data + regression test =
Logistic Regression
Ordinal data + Regression test =
Multinominal Logistic Regression
Interval data + Regression test =
Linear Regression
Logistic Regression
multiple independent variables used to predict outcome/dependent variable (DV)
predicted outcome data level = nominal
Multinominal Logistic Regression
multiple independent variables used to predict outcome/dependent variable (DV)
predicted outcome data level = ordinal
Linear Regression
multiple independent variables used to predict outcome/dependent variable (DV)
predicted outcome variable data level = interval
List the regression test for each data level.
Nominal, ordinal, and interval data
logistic regression
multinominal logistic regression
linear regression
For Pearson Correlation test, what does a non-significant value indicate?
(p > 0.05)
No LINEAR correlation between groups; however, may be NON-LINEAR correlations present
delineate correlation vs. regression types of statistical tests
Correlation: relationship between TWO groups; fixed value range
Regression: relationship MULTIPLE IV in predicting an outcome variable (DV);
what type of statistical test will only produce values within [-1 - +1]?
Correlation
Partial Correlation
Correlation test ONLY for interval data (Pearson Correlation) that controls for confounding variables
List all statistical tests that can be represented by Kaplan-Meier Curve
Log-Rank test
Cox-Proportional Hazards test
Kaplan-Meier test
All survival tests
“over time” indicates what type statistical comparison
survival test
Survival test =
Compares proportion of events over time, or time-to events, between groups ~ongoing progression~
“free from an event” over time
survival
Log-Rank test
survival test + nominal level data
died/didn’t die or developed disease/didn’t develop disease
Cox-Proportional test
survival test + ORDINAL level data collected
ex. - tumor growth - 0 in., 1-5 in., 5-10 in., etc.
Kaplan-Meier TEST
survival test + INTERVAL level data
List all survival statistical tests for each data level:
nominal = Log-rank ordinal = cox-proportional hazards test interval = kaplan -meier test
What is the most commonly utilized type of comparison of statistical data?
Group comparisons
Which Correlation test can control for confounding factors?
Partial Correlation of Pearson Correlation test
What kinds of comparisons are made in group comparisons?
frequencies
counts
proportions
Type of statistical test assess event-occurrence and time-to-event?
survival tests
What data being compared in Paired/Related Group Comparisons?
Comparing data of subjects from the SAME study groups
“pre- vs. post-“ “before vs. after” “baseline vs. end” should all indicate what kind of data?
Paired or related data - group comparisons
What are the required assumptions of interval/ratio data for proper selection of parametric test?
- Normally-distributed
- Equal Variances
- Randomly-derived + Independent
How do you handle interval/ratio data NOT normally-distributed?
1) TRANSFORM data –> Standardized Values (z-score/Log transformation) which might be Normally-distributed
2) Use Non-Parametric tests –> use equivalent stat test of the lower data level
* Always run descriptive statistics and graphs*
Null Hypothesis
Research perspective that states there will be NO true differences between groups being compared
p value of statistical tests is what is used to determine ACCEPT or REJECT null hypothesis
Mann-Whitney test
Ordinal data
2-groups
Independent data
compares MEDIANS
Wilcoxon Signed Rank test
Ordinal data
2-group
Paired/Related data
compares MEDIANS
Kruskal-Wallis test
Ordinal data
** 3+ groups **
Independent data
compares MEDIANS
** If p value < 0.05, run a POST-HOC test to determine between which groups the difference exists!
Friedman test
Ordinal data
** 3+ groups **
Paired/Related data
compares MEDIANS
**If p value < 0.05, run a POST-HOC test to determine between which groups the difference exists!
Name Post-Hoc tests for ORDINAL data:
Student-Newman-Keul
Dunnett
Dunn
Kruskal-Wallis and Friedman
What type of tests utilize Post-Hoc tests?
Group Comparisons
3+ groups
p value < 0.05
Ordinal data stat tests:
- Kruskal-Wallis test
- Friedman test
Interval data stat tests:
- ANOVA/ANCOVA
- Repeated Measures ANOVA/ANCOVA
(Pearson’s) Chi-Square test
Nominal data
2-groups
Independent data
compares PROPORTIONS
