Statistical tests Flashcards
One sample T-test
- used to compare mean of a single sample with a known population mean to check if the sample is derived from the described populations
- needs two values- the observed mean and the population mean
- variable must be normally distributed
- sample size must be adequate enough to prevent extreme skewness
Two sample t test
- Student’s t
- used to compare means of two samples
- needs independent variable and dependent variable
- unpaired t test compared independent smaples
- paired t test compare the same group e.g pre and post test
- t-test assumes equal variance among the two groups of given data (use Levene’s test to check)
ANOVA
- used to compare means of multiple groups
- based on variance comparison
- one way ANOVA is used for one independent variable compared across more than 2 groups
- 2 way ANOVE has 2 independent variables
- repeated measures ANOVE is used when the same measure is used on multiple occasions ie. paired observations
- ANOVA is calculated using the ratio of variation between groups to the variation within groups- F statistics
Disadvantage of ANOVA
- can only tell us if a significant difference exists among groups but doesnt say where the difference comes from
- assumes normal distribution and equal variance
1 mean
One sample t test
Two means- unpaired
Two sample t test
Two means paired
Paired T test
More than two unpaired means
One way ANOVA
More than two paired means
MANOVA
One Median
Sign test
Two Medians unpaired
Mann-Whitney U test
Two medians paired
Wilcoxon rank sum test
More than two groups of unpaired medians
Kruskal Wallis test
More than two groups of paired medians
Friedman test
Two proportions- unpaired
Chi Square test
Two proportions paired
McNemar test
More than two proportions unpaired
Log linear or logistic regression
Parametric tests
- used if at least one variable is quantitative and normally distributed
- t-tests or ANOVAs
Non-parametric tests
- used when a) both dependent and independent variables are qualitative (nominal or ordinal) or b) when the variables are quantitative but not normally distributed
- ranks are compared
- sign test, wilcoxon rank sum, mann-whitney U, Kruskal wallis
Sign test
-simple non-parametric test that compares the median of a sample to the median of the population
Wilcoxon rank sum test
-used to compare the two paired observations in non-parametric fashion
Mann-Whitney U
-used for independent observations in two groups
Kruskal-Wallis
-used for 3 or more groups parametrically
Transformation of data
-log transformation is used to convert data into a form more acceptable for parametric analysis
Log transformation
- most common transformation of data
- in right skew this yields a normal distribution (lognormal curve)
Square root transformation
- has normalising and linearising properties
- stabilises variance
- normalises Poisson distributions
Reciprocal transformation
-used in survival rates
Logit transformation
- used in the distribuution of proportions
- linearises a sigmoid curve
Chi square x2 test
- commonly used non-parametric test
- used for comparing frequency counts or proportions
- contingency table is made like a 2x2 table
- observed frequencies vs expected outcomes (if null hypothesis was true)
- ration between observed to expected frequencies in the cells of the table is the chi square
Fisher’s exact test
-used in place of chi-square if the expected cell frequencies in more than 20% of cells falls less than 5
Yates correction
-used if the total sample was less than 100 or any cell was less than 10 in chi squared
McNemar test
-X2 test for paired data
Mantel-Haenszel test
-form of chi square test wherein the influence of two dichotomous categorical variables on one dependent variable is tested
Log-linear analysis
- used in chi-square test where log-values of cell frequencies are employed
- this is when more than 2 groups or variable are studied