1/19
Looks like no tags are added yet.
Properties of a normal distribution
continuous, symmetrical, ~68% of the area is within 1 standard deviation of the mean, and ~95% is within 2 standard deviations. Mean, median, and mode are equal
Properties of standard normal distribution
Standard deviation is always 1, and the mean is always zero
Central limit theorem
The sum or mean of a large number of measurements randomly sampled from a non-normal population is approximately normally distributed
3-prong approach
Histogram, qqplot, Shapiro-Wilk test
Log transformation
Used when measurements are ratios or products of variables, the skew is to the right, the group with the larger mean also has a larger standard deviation, and when the data spans orders of magnitude
Arcsine transformation
P is a proportion (percentages must be converted to proportions)
Square-root transformation
Used when data are counts, can be used to equalize standard deviation
How to decide id a test is appropriate
Check data type and test assumptions
Binomial test
Data type is proportion, test assumption is random sampling. Null distribution allows us to see how probable our distribution is if the null hypothesis is true
Chi-square GOF test
Data type is frequencies, assumptions are that there are no categories with an expected frequency less than one and no more than 20% of categories should have an expected frequency of less than five. Degrees of freedom is # of categories minus 1
Chi-square contingency test
Data type is frequencies and test assumptions are the came as GOF test; compares two categorical variables to see whether they are independent. Degrees of freedom is (r-1)(c-1)
t-tests
2 continuous numerical variables, calculate the difference between the sample mean and the true mean, divided by the estimated standard error. CI’s are less precise, so a greater difference between means is required to achieve statistical significance
1-sample t-test
We are comparing our sample mean to a hypothesized population mean (null). AssumptionsL random sampling, variance is normally distributed. df is n-1
Paired t-test
We are finding the difference in means between paired measurements, then comparing that to the hypothesized difference of zero
Two-sample t-test
We are comparing two means to decide whether the difference is actually zero (null). Assumptions: random sampling, normal distribution, standard deviation and and variance are the same in both populations
How do we know if variance is equal?
Levene’s test for homogeneity of variance, which tests for heteroskedasticity
Non-parametric tests
Used when we’ve violated our assumptions, because the only one they have is random sampling. They have less power than other tests, so its harder to reject the null
Sign test
We compare ranked measures against either hypothesized values of other ranked measures, and we expect half of our data to fall above the hypothesized median. We test this with a binomial test
Mann-Whitney U test
Compares two independent samples; after we rank all the data, we find the rank-sum of each group, and the number of times a data point is smaller than those of the other group
Permutation test
Breaks up the association between variables in order to create a null distribution, which is then compared against the real test statistic