Chi-square test in SPSS + interpretation (assumptions violated)

okay let's have a look at a chi-squared test when the assumptions have been violated I'm going to do two examples one is for a two by two table and one is for a table that's bigger than two by two now the advice I'm going to give in this video is from a couple of stats books one is Andy fields discovering statistics using SPSS his recommendation and the other is one by Julie pelant now if your lecturer tells you something different please do use their advice and what they want you to use but this is just standard in most text okay let's look at our two by two example first if we go to analyze descriptives and chi-square or cross tabs just as we did before and we're going to look at gender and smoke smoking status under statistics choose chi-squared and the effect size Phi will be the one we want to look at for a two by two continue under cells we want to take expected because this is how we're going to check whether or not the assumptions have been violated click continue and then click okay I'll write in our output here's our crosstabs table now the assumption for a two by two table meaning two categories by two categories is that all the expected counts are at least ten now you can see that two of them here are one is seven point eight one is eight point two Bulls less than ten which means we violated that assumption when that happens with a two-by-two table we go down here we can use Fisher's exact test the likelihood ratio is also an option but with the table with just a two by two table Fisher's exact test is is common to use I'm just going to use the two-sided figure here and that's when your alternate hypothesis is two-tailed in other words you don't specify a direction of the difference for example you don't say there's more male smokers or more male non-smokers you just say there is an association between gender and smoking status once you determine which significance value is best to use according to your alternate hypotheses you want to compare it to your alpha just as usual now alpha which is our level significance is typically 0.05 we can see that this is much bigger than 0.05 which means that a result is not significant and we do accept the null hypothesis also known as h0 our null hypothesis and a chi-square test is always there is not a significant difference or a not a significant association between the two variables in other words smoking status is completely independent of someone's gender so that is the result that we would accept in this case because this is not significant and when you get a non significant result you do not look down here at the effect size because it's it doesn't matter because the result is not significant now let's have a look at an example with a table bigger than 2x2 here I have some insurance data and the two variables we're going to use are the type of claim someone has filed and whether or not that claim turned out to be fraudulent so let's conduct our chi-square test is normal analyzed descriptives crosstabs type of claim because this has more categories I'm going to make it my row variable and fraudulent is only two it's yes or no so I'm going to make it my column just so my tables easier to read statistics chi-squared and this time we're going to use Cramer's V continue and in cells I'm going to click expected but it's not absolutely necessary this time click continue and ok here's my crosstabs table it's a bit bigger this time it's bigger than a 2x2 and our assumption for a table bigger than 2x2 is that the expected count is not less than 5 or 20 percent of the cells have expected count greater than 5 so if we look down here at the bottom it says 6 cells or 60% have expected count less than 5 this violates the assumption because 60% is much bigger than 20% so I'll say it again we want this value to be 20% or otherwise the assumption has been violated if you need to rewind and listen to that again go ahead I know it's a lot to take in and one go when that's the case we're going to read off the likelihood ratio here's my statistic my degrees of freedom and my significance value again I'm going to compare it to my level of significant level significance 0.05 this is much bigger which means I accept my null and I conclude that there is no association between the type of claim and whether or not the claim is fraudulent

Channel: BrunelASK

Views: 317,307

Rating: 4.9161072 out of 5

Keywords: Chi-squared Test, SPSS (Software), Tutorial, Assumptions violated, Likelihood ratio, Fishers Exact, SPSS v20

Id: t7FMonySjDY

Length: 5min 2sec (302 seconds)

Published: Tue Aug 13 2013