LECTURE 5 Flashcards
Parametric stats are used to analyze ____data
quantitative
true/false: data needs to be normalized for parametric stats
true
based on t, F, chi-square distribution
-test, ANOVA, Pearson correlation, linear regression
examples of
Parametric statistics
Spearman rho, Mann-Whitney U, Friedman’s ANOVA, Wilcoxon-signed ranks
examples of…
non-parametric stats
This is our option when we have violated assumptions, or we have nominal or ordinal data.
non-parametric stats
Parametric Assumptions for t-test/one way ANOVA
- IR data
- normality
- homogeneity of variance
- free of extreme outliers
- independence of observations
if you knew that the population is normally distributed…even small samples (n<30) will meet this assumption.
In practical terms, as long as your sample is fairly large, outliers are a much more pressing concern than ______
normality! (need 30 bc of CLT)
how do you check for normality?
histograms
look at skew/kurtosis (greater than 2 or smaller than -2)
shapiro-wilk test (want it to be non-significant at 0.05-greater than 0.05)
for skewness and kurtosis, we are looking for
Is data more than 2, less than -2?
IF SO, NOT NORMAL
How would you test for HOV?
Levene’s Test
What is Levene’s Test?
Tests if variances in different groups are the same.
Want this test to be not significant. You want them to show “no difference.”
Set alpha at .05.
In t-test/ANOVA, what will influential outlier do to data?
pull mean to outlier
For regression tests, what will influential outliers do?
pull best fit line towards outliers
How can you look for influential outliers?
- histograms
- skewness/kurtosis
- boxplots
- regression: COOK’S DISTANCE GREATER THAN 1
-Data must be Independent
-scores must not follow a pattern over time
-scores from one participant can’t influence another participant’s scores.
What assumption for parametric stats is being met?
independence of observations
(scores shouldn’t trend over time
data can’t be from same body parts of same person…)
regression assumptions
- linearity
- homoscedasticity
- outlier testing in regression
In correlational/relationship analyses (ex/ regression), the variance of the outcome variable must be about the same at all levels of the predictor variable.
If the variance is not evenly distributed, it’s called having
heteroscedasticity