Choosing Statistical Test 1 Flashcards
What must one ask when choosing a statistical test?
What is being tested? Are we comparing different samples for a point estimate (mean, median, frequency) - category 1
or are we demonstrating associations and relationships? - category 2
For category 1 questions, what must we ask?
What is the nature of the point estimate - mean or proportional?
How many groups are being compared?
How many observations of similar nature are made in each subject - paired or unpaired (independent) observations?
Can we assume parametric distribution - is the data not continuous?
Statistical tests for means
Parametric
Statistic test for means of one group
One sample test
Statistical test for means of two unpaired groups
Two sample test
Statistical test of means of two groups which are paired?
Paired t test
Statistic test for means with more than 2 groups which are unpaired
One way ANOVA
Statistical test for means of more than 2 groups which are paired
MANOVA
Repeated measures ANOVA
Statistical test for medians
Nonparametrical
Statistical test for medians of one group
Sign test
Statistical test for medians for two groups which are unpaired
Mann-Whitney U test
Statistical test for medians of two groups which are paired
Wilcoxon rank sum test
Statsistical test for medians of more than 2 groups which are unpaired
Kurska-Wallis test
Statistical test for medians of more than 2 groups which are paired
Friedman test
Statistical tests for proportions
Categorical
Statistical tests for proportions of 2 groups which are unpaired
Chi square test
Statistical test for proportions of two groups which are paired
McNemar test
Statistical test for proportions of more than 2 groups which are unpaired
Log-linear or logistic regression
What can parametric tests be used?
If at least one variable is quantitative and normally distributed
Which parametric tests are used when comparing groups?
T-tests
ANOVA
When are non-parametric tests used?
When either both dependent and independent variables are qualitative (nominal or ordinal)
or when variables are quantitative but not normally distributed
What happens in nonparametric tests?
Ranks are compared
What is the sign test?
Non-parametric test that compares median of a sample to median of the population
What is the non-parametric equivalent of the one sample t test?
Sign test
What is the Wilcoxon rank sum test used for?
Comparing two paired observations in nonparametric fashion
What is Wilcoxon ranks um test equivalent to?
Paired t test
What is Mann Whitney U test used for?
Independent observations in two groups
What is Kruskal-Wallis
Non-parametric one-way ANOVA equivalent used for 3 or more groups
Which statistical tests offer more robust analysis
Parametric
Most commonly used method to transformation data from non-parametric to parametric?
Log transformation
Square root transformation
Square transformation
Reciprocal transformation
Logit transformation
For what type of results does log transformation yield a normal distribution?
Right skew (positive)
What is another name of normal distribution using log transformation?
Lognormal curve
What can log transformation for right skew (positive) results do?
Yield normal distribution
Linearize a nonlinear correlation with upward shift
What does square root transformation do?
Normalise and linearize
Stabilize variation
What is square root transformation used for?
Normalising Poisson distributions
What is reciprocal transformatino often used for?
Analysis of survival rates
What is logit transformation often used for?
Distribution of proportions
What does logit transformation do?
Linearizes a sigmoid curve
CI may be difficult to interpret after transformation
Geometric mean values may get distorted
What is chi square test commonly used for?
Comparing frequency counts or proportions
What type of test is Chi squared?
Non-parametric
How does a Chi square test work?
Contingency table is constructed
Rows and columns are mutually exclusive
What is a contingency table?
A table whose contents are in frequencies
Classifications in contingency tests of chi square test
Observed frequencies = real outcomes observed at conclusion of study
Expected frequencies = outcomes predicted if null hypothesis was true
What is chi-square?
Ratio of observed to expected frequency
When is fisher exact test used in place of chi square?
If expected cell frequencies in more than 20% of cells falls less than 5
When is Yates correction used in chi square?
If total sample is less than 100 of any cell value is less than 10
When must chi square always be used?
Only for original frequency counts and not percentages
What is the chi square test for paired data?
McNemars test
What is the Mantel-Haenszel test?
A form of chi square test where the influence of binary categorical variables on one dependent variable is tested
What is log-linear analysis?
Form of chi square test where log values of cell frequencies are employed
When is log-linear analysis used?
When more than 2 groups or variables are studied
Questions to ask if using statistical tests for category 2 data
How many dependent variables?
How many independent variables?
Are the variables categorical or continuous?
Statistical tests for category 2 data
Testing association
Test used if one dependent & independent variable, testing correlation in continuous, parametric data
Pearsons
Test used if one dependent & independent variable testing correlation in non-parametric data
Spearmans
Kendall’s tau
Test used if one independent & dependent variable testing regression in continuous data
Simple linear regression
Test used if one independent & dependent variable testing regression using categorical data
Simple logistic regression
Test used if one dependent variable and >1 independent variable testing regression in continuous data
Multiple linear regression
ANCOA
Test used if one dependent variable and >1 independent variable testing regression in categorical data
Multiple logistic regression
Test used if 2 or more dependent variables and >1 independent variable testing regression with categorical data
MANOVA
Test used if two or more dependent variable and >1 independent variable testing regression using continuous data
MANCOVA
Multivariate multiple regression
