10. Exploring Assumptions Flashcards
steps to conducting statistical analyses
- explore your data
- check assumptions
- conduct statistical tests
How do you explore your data?
- graphs
- run descriptive stats
types of data
parametric
nonparametric
assumptions about parametric data
- normally distributed
- homogeneity of variance
- at least interval data
- independence
independent samples
data from different subjects are independent
dependent samples
behavior of one subject doesn’t influence behavior of another
What does it mean by normally distributed?
- sampling distribution and sample data are both normally distributed
- central limite theorem
How to assess normality
- visually via graphs
- descriptive statistics
- comparison to normal distribution and assess for differences
two graphs to use to assess normality
- histograms
- p-p plots
How might histograms be useful for assessing normality?
- frequency distribution
- can add a normal distribution overlay
What is a p-p plot?
probability-probability plot
What does a p-p plot do?
- plots probability of a variable against the probability of a normal distribution
What does a p-p plot convert scores to? Why?
z-scores
to compare against z-scores of normally distributed data
data from the sample
actual/observed probability
normally distributed data
expected probability
What sort of descriptive stats would you use to assess normality?
- measures of central tendency
- measures of variability
- measures of shape
What are the measures of shape used to assess normality?
skewness
kurtosis
How do you compare data to a normal distribution?
2 tests can be used to determine
- Kilmogorov-Smirnov test
- Shapiro-Wilk test
What is the benefit of the Shapiro-Wilk test?
more power to detect differences from normality
With tests that compare to normal distribution (Kolmogorov-Smirnov and Shapiro-Wilk), what does P > 0.05 mean?
indicates that there’s no difference between the sample distribution and normal
What are the limitations to tests that compare to normal distribution?
- not always accurate with large samples
- small changes can lead to significant test results
What must you always do in addition to running tests?
graph the data
What does homogeneity of variance mean?
the spread of scores around the mean should be similar in each group
What type of design does homogeneity of variance apply to?
non-repeated measures designs
How to test homogeneity of variance
- correlation
- comparison of means
homogeneity of variance: correlation
uses graphs
What test is used to test for homogeneity of variance?
Levene’s test
What does Levene’s test assess?
assesses the null hypothesis that variances in different groups are equal
For Levene’s test, what does P less than 0.05 mean?
- variances are different among groups
- assumptions have been violated
What are limitations to assessing for homogeneity of variance using Levene’s test?
- subject to bias with large sample sizes
- small deviations produce significant Levene’s test with large samples
Ways to deal with outliers
- remove the case
- transform the data
- change the score
dealing with outliers: removing the case
- delete the data
- only done if there’s a good reason to believe it’s not from the population you intended to sample
dealing with outliers: transforming the data
reduces skewness
dealing with outliers: change the score
can be used if transforming data fails to normalize the distribution