Non Parametric Statistics Using Stata

[Music] what I'd like to do now is demonstrate how to conduct several nonparametric analyses using Stata and to do this I'm going to use the growth and enhancement data for Montana students and the data that I'll be using to illustrate these analysis is gender high school GPA criterion reference to 4th grade math scores high school GPA for two different classes class 1 in class 2 and then pretest reading in post-test reading the first analysis that I'd like to demonstrate is comparing gender on high school GPA now remember we use nonparametric statistics when the assumptions of normality and homogeneity of variances are violated and in the case of correlations where the assumption of a linear relationship is violated so when we compare gender male and female on high school GPA typically we would use a independent samples t-test but for a nonparametric analysis what we're going to do is we're going to use I'll go to summary tables and tests and nonparametric will use a mann-whitney u test or in Stata the equivalent is the wilcoxon rank-sum test and so this is a very easy test to run our variable in this case is high school GPA and our grouping variable is gender and then we'll submit that and we'll toggle back to our output and we can see that the ranked sums the sum of the ranks for males is twenty-five thousand six hundred eighty five and for females is 19 thousand four hundred and sixty four so this difference is significant and can see here that our probability as much greater or much smaller I should say than point zero five so males have significantly higher GPA and females now the next analysis I want to show you is the case where we have a pretest and a post-test so in this case I have pretest reading and post-test reading scores and normally we would use a paired or a dependent samples t-test but in a nonparametric world we can go to summaries and tables and tests and nonparametric tests of hyppolit hypotheses and we'll use the wilcoxon matched pairs sign rank test and so our variable in this case is pretest and our comparison variable is post-test reading and so this is the variable that's being compared pretest and post-test reading and so we submit that and we see that the sum of the positive ranks compared to the some of the negative ranks or the difference between pre and post-tests is much greater and so our probability of significance we do have a significant result in this case point zero zero zero so again this is very much less than point zero five now the only thing is to get a reference to see how males and females differ is we like to see what the means are so we're gonna have to go up to get our summary statistics and we can put in pretest and post-test reading and then we can see that the mean for pretest was 282 and post-tests was 290 so this difference between the pre and the post test was significant now a third type of nonparametric statistic that you might be interested in conducting is when the assumptions for the ANOVA analysis are violated so again we have a procedure ik one-way analysis of variance and so we can go to our nonparametric test of hypotheses and we go to the kruskal-wallis ranked test that's the equivalent or the nonparametric equivalent of the one-way analysis of variance and so in this case what I want to do is compare class there's four classes and I want to compare them on high school GPA and so when I come back here to my to my window to do the high school do the comparison my outcome variable is high school GPA and my groups will be class and so when I run that analysis I can see then I get a output that gives me the ranked sum for each of the classes and what this is telling me is that the probability that there's one class that's significantly different from another class is not significant because this value is greater than 0.2 nine to seven so that's the kruskal-wallis for when you have when you want to compare more than two group means and use a one-way ANOVA but the assumptions for the parametric ANOVA are violated the last procedure I want to demonstrate is when we have rank data and I need to open up another dataset for that and so let's see if I can find that okay so in this situation in this data set I have two judges and what they've done is they've ranked students from high to low so they've ranked them from the highest performance to the lowest performance in 4a for their science fair projects and so we're gonna compare the two judges to see how the rankings compare in terms of how consistent these judges are in comparing these students now normally you might consider doing a Pearson correlation but in this case since we have rank data we're going to use a nonparametric correlation so we can go to statistics again summary tables and tests nonparametric and then we can go to the Spearman correlation for rank data and so in this case we simply put in judge one put in judge two we like to have the correlation coefficient the number of observations and the significance level and then we go okay and we can see that our correlation between judge one and judge two there's ten observations and the Spearman correlation is point eight two so that's a large correlation so these judges are fairly consistent in terms of the way that they rank students in terms of their performance on their science fair projects well this was just a short little tutorial on running nonparametric statistics and Stata and I hope it's helped

Channel: Arthur Bangert

Views: 8,961

Rating: 4.9230771 out of 5

Keywords:

Id: O88m9iixhTY

Length: 8min 42sec (522 seconds)

Published: Mon Jul 16 2018