Chi-square Test: a test for frequency tables with categorical data
o 2 cases: one-variable case/Goodness-of-fit test two-variable case/Test of Independence o not about the mean of the entire data set, but about the expected frequency When we are making hypotheses about frequencies, we have to remember: o We either reject or retain our null hypothesis We retain if the test statistic is extreme or more extreme if the null hypothesis were true We reject if we are unlikely to obtain a test statistic as extreme or more extreme if the null hypothesis were true Pearsons Chi-Square Test o Frequency(observed)-f(expected)^2/frequency(expected) o If we have a series of observations falling into different categories with given probabilities, we can know approximately what the sampling distribution of the following test statistic o Df=number of cells 1 o We compare the observed vs. expected X^2 statistic o Same steps as that of a hypothesis test: Specify null and alternative hypothesis Draw a random sample: calculate your obtained and expected frequencies Usually we expect the same frequency in all cells Calculate X^2 value Compared your calculated X^2 value to the critical X^2 value (based on our df and alpha) and make a decision about the null hypothesis Assumptions and Limitations: o Should be a random sample from population o Observations should be independent o Assumes that frequencies are normally distributed Can also use the Chi-squared test to see whether 2 categorical variable are independent or related o In order to get the expected frequency: Frequency(expected)=(row sum)(column sum)/total Then we can use the same formula as stated earlier to calculate our Chi-square statistic Things to Note: o X^2 test are never negative X^2 test is always positive UNLESS your expected and observed frequencies coincide perfectly Remember that the difference between the t and z statistic is that there is no measure or estimate of population variance with the t statistic Effect Sizes and Chi-Square o Omega for Goodness of Fit (only for one variable) Varies from 0-1 Small effect: 0.1 Moderate effect: 0.3 Large effect: 0.5 But you will never actually need to calculate omega in this class