Logistic Regression - SPSS (part 1)

hi welcome to how to stats comm in this video I'm going to demonstrate how to perform a binary logistic regression so binary logistic regression is when you have one dependent variable that is dichotomous so there are only two categories that you're trying to predict now logistic regression is a appropriate for that type of analysis and and it also accommodates independent variables that are scaled on an interval ratio scale or an ordinal scale or a nominal scale and in this fictitious example I've got a dependent variable that is dichotomous and it's credit default and what I've simulated our data that are now let's just talk about I'm talking about subprime mortgages whose going to default on a subprime mortgage so I've got my zeros that demarcate people who did not default and then I've got a series of ones that are interspersed across the data I've actually got the data sorted across set a salary to a certain degree so it does look like it's organized so there's there are some ones in there but there's far fewer in frequency in fact it's about 10 percent of people that I've simulated that are defaulting on a subprime mortgage that's probably higher than average but I just use that as an example and my independent variables are annual salary and gender and so I'm going to try to use these two independent variables to predict who is going to default on a subprime mortgage and so salaries measured in thousands of dollars so this person here earned $120,000 and gender zero one one is male and zero is female I believe there's about 50 percent equal here of males and females alright so now to perform the binary logistic regression going to analyze regression binary logistic you got credit default which is the dependent variable annual salary and gender I'm going to put that into my covariance these are basically the predictors the their actually that's exactly what they are other predictors now because gender is dichotomous only so it's only got two levels in the nominal variable I can actually include it ferrets but if you had another variable that had more categories into it actually have to use the categorical option here and I'm not going to talk about that in this video series but it does add complexity to your analysis and the output so I'm trying to keep this relatively simple I'm going to use method enter so I'm going to force both independent variables into the model I got block 1 of 1 so I'm not actually going to do a hierarchical logistic regression which would be a little bit more complicated I'm just entering them into one block and I'm forcing them both into the equation there is one option I want to choose which is Hosmer lemons how goodness-of-fit the rest are I would argue probably a little bit less important I'm not saying you can't get valuable information from probably the 95% confidence interval so probably fairly interesting but I'm going to try to keep it letter to be simple because logistic regression is relatively complicated already I am going to save two variables probabilities and group membership and if I have enough time I'm going to go through that my hunches as I won't because I'm going to try to keep this just to the to the basics but if I feel like I've got time to do so I'll actually talk about that in more detail than what you get information from the output all right so that's I'm going to click off those options I'm going to click OK ok so the first output table is telling me that 189 people in the sample no missing data then is saying my dependent variable coded no yes as in no did not default on their subprime loan and yes did default on their subprime loan and then the next table which is after block 0 now block 0 means that there's no there are no predictive variables included in the in the model it's really just a intercept only model it's basically a null model if you will so it's important now it's actually quite important to still interpret the null model I think a lot of people just blow by it and and I think you don't get any interesting information it looks like SPSS is actually crashing on me I can't actually move my let me see if I can get that back up okay there we go okay I've jumped down a bit okay so here's a classification table now this is a null model with no prediction and the classification accuracy here is 90.5% now that sounds impressive it's not models basically saying what if we predicted everyone as a non-default what accuracy would we get well because the vast majority of people don't

Channel: how2stats

Views: 520,253

Rating: 4.6602149 out of 5

Keywords: logistic regression, SPSS, credit default, tutorial

Id: OvQShzJ7Sns

Length: 5min 3sec (303 seconds)

Published: Thu Aug 04 2011