Pearson Chi-Square, Continuity Correction, and Fisher's Exact Test in SPSS

oh this is dr. grande welcome to my video on interpreting and Chi square the continuity correction and Fisher's exact test in SPSS as always if you find this video useful please like it and subscribe to my channel I certainly appreciate it I have here in the SPSS data editor fictitious data I'll be using for this example I have two variables training and outcome training has two levels online and face-to-face and outcome has two levels fail and pass so let's assume that we have a specialized training program for counselors a program design to teach a specific skill and we offer it online and face to face and then we have an assessment at the end of the course to determine whether the counselor has learned the particular skill that we are trying to teach so the outcome is fail or pass so training site dichotomous variable outcome is a dichotomous variable as well so this is a 2x2 chi square that we'll be conducting so what does the chi-square test tell us well in this case it's a chi-square test of Independence and it tells us if the training variable is independent of the outcome variable it doesn't speak to causality meaning online versus face-to-face and fail versus pass we're not saying that training causes the outcome rather is the training variable independent of outcome and the null hypothesis in this case is that these variables are independent so to conduct chi-square I'm going to go to analyze descriptive statistics and then crosstabs and here we look at the research design even though the chi-square results won't tell us about causality we are assuming that training we're treating training has an independent variable and outcome as a dependent variable so typically in this type of situation training we go in the row list box and outcome in the column list box it can be done the other way around how it's just a tradition to put the independent variable or the one we're treating as an independent variable in the robust box and of course the dependent variable in the column list box under statistics just going to add chi-square going to check that off I'm not going to check off any other of the options here it could continue under cells observed it is checked off by default four counts I'm going to add the expected counts and the row column and total percentages click continue no changes under format or style and click OK to conduct chi-square we have the output here in the statistics viewer and if we look at this training times outcome cross tabulation there are a few things we can look for here in this example if we look at online versus face-to-face in terms of the count we can see that for online we have 24 fail and 21 pass but for face-to-face it's 14 fail and 31 pass so before even looking at the results of any of the tests we are led to believe here that face to face is more associated with passing now whether it's statistically significant is something we have to look at in the chi-square tests but here we can see the counts we have 21 pass for online compared to 31 face to face and of course the difference in ten with the fail level of outcome as well 24 to 14 another thing we want to pay attention to here are the expected counts and we can see the lowest expected count is 19 so if we move down to the chi-square tests notice here at the bottom 0 cells have expected count less than 5 the minimum expected counts 19 again we saw that up here in the training times outcome cross tabulation so in looking at the chi-square tests table normally here in the situation we would interpret the Pearson chi-square because we have no cells that have an expected count less than 5 and the minimum expected count is 19 also in this example our sample size is greater than 40 so the sample size being greater than 40 and zero cells having expected count less than 5 it's not unusual to interpret Pearson chi-square and in this case point 0 3 3 that is statistically significant so we would say that e not have independence between training and outcome they are not independent the two variables are not independent of one another point 0 3 3 less than the alpha of 0.05 however there are a lot of guidelines on this issue in terms of the yates correction which is this continuity correction and Fisher's exact test so one of the popular guidelines for the continuity correction is if we have an expected count less than 10 in any of the cells so we look up here to the cross tabulation and we have a value less than 10 we'd use the continuity correction Yates correction a popular guideline for interpreting Fisher's exact test would be that if any of this have an expected count less than five so less than ten for continuity correction less than five for Fisher's exact test another guideline we see with Fisher's exact test is that if greater than 20% of the cells have an expected count less than five so there is not agreement on this issue of Pearson chi-square versus the continuity correction and Fisher's exact test and even though in this case you can see the results are very close to one other the p-values are very close to another we would have a different finding so we would assume the two variables are not independent based on Pearson chi-square value point the p-value of 0.03 three and we would say they are independent we would fail to reject the null hypothesis and say the variables are independent with the continuity correction point zero five five and the result the p-value resulting from Fisher's exact test point zero five four so with all these different guidelines and theories regarding these three tests which one do we use how do we interpret the chi-square test table I think that when you have a Pearson chi-square and you do have that sample size greater than 40 and zero cells have expected count less than five a good argument could be made for using the Pearson chi-square and if any of the cells have an expected count less than five or you have sample size less than 40 then Fisher's exact test would be a good choice and in that instance there is no real need for the continuity correction meaning using those guidelines we wouldn't use Yates correction we would just be deciding between Pearson chi-square and Fisher's exact test remember that no matter which set of guidelines you use in terms of interpreting these different tests set that decision in place prior to running the analysis so that you know coming in that based on a certain expected count or sample size or both you'll be interpreting one statistic instead of another I hope you found this video on the chi-square test and SPSS to be helpful as always if you have any questions or concerns feel free to contact me I'll be happy to assist you

Channel: Dr. Todd Grande

Views: 52,730

Rating: 4.9185061 out of 5

Keywords: SPSS, chi-square, 2 by 2, Fisher’s Exact Test, Fisher’s Exact, Fisher’s Test, Fisher, Fisher’s, Exact, Test, Pearson Chi-Square, continuity correction, Yates, Yates’ Correction, crosstabs, crosstabulation, Chi-Square Test of Independence, independent variables, predictor variables, dependent variables, outcome variables, dichotomous variable, dichotomous, binary, p value, data, analysis, statistics, counseling, Grande

Id: YYw9fZq1jEQ

Length: 9min 14sec (554 seconds)

Published: Mon Dec 19 2016