Binary Logisitic Regression in SPSS with Two Dichotomous Predictor Variables

oh this is dr. Grande welcome to my video on performing a binary logistic regression in SPSS in this example I'll be using two dichotomous predictor variables so taking a look at these fictitious data have loaded the data editor in SPSS you see I have four variables one's an ID I have a hundred records in the state asset I have an independent variable or predictor variable gender an independent variable referral and a dependent variable outcome and notice that all these are nominal and where precisely they're all dichotomous so if I click a one up here you can see that it's all zeros and ones so each variable is dichotomous you have male and female voluntary and involuntary and unsuccessful and successful so let's assume these data were collected from a program designed to treat substance use and all the participants that go through the program fall into two categories successful or unsuccessful so either an individual is able to discontinue using substances which case we have successful or they were unable to discontinue use and in that case we have unsuccessful all the outcomes can be placed in either the successful or unsuccessful levels of the variable outcome so these data and the research question that we are asking or ideally setup for a binary logistic regression in a binary logistic regression you have one dependent variable and it has to be dichotomous and that's what we have in this case outcome as successful or unsuccessful and then you could have multiple predictor variables and those predictor variables can be nominal ordinal interval or ratio in this case I have - predictor variables - independent variables and they are both nominal and more precisely dichotomous if you had more than two levels for your independent variables that could still work and if you had interval or ratio level data which SPSS refers to a scale that would work an ordinal would work as well so what a logistic regression does is it gives you the probability that a certain combination of independent variables will have a particular outcome and as you can see in this example there are four possible combinations you have female and it's an involuntary referral male with an involuntary referral a female with a voluntary referral to counseling services and a male with a voluntary referral to the counseling services because of the nature of the output of logistic regression I find it useful to run a chi-square first just to get an idea of what you may expect to find in the logistic regression results so I'm going to go to analyze descriptive statistics and then cross tabs for rows I'm going to put the independent variables or predictor variables so it's going to be gender and referral and then for column I'm going to put outcome under statistics I'm going to check off chi-square press Continue and then under cells observed counts are checked off by default I'm going to add expected and percentages for row column and total and click continue and then click OK so first we have the crosstab gender x outcome as you can see if we look at male and female looking at the numbers for the male participants we have 34 successful in the substance use treatment counseling program and 31 unsuccessful for the female participants 29 successful and six unsuccessful so before we even get to interpreting the results of logistic regression we know that the probability of a male being unsuccessful in this treatment program is going to be higher than the probability of a female being unsuccessful and similarly we want to take a look at the crosstab table for referral times outcome we can see we have the voluntary and then the involuntary and again we can see here looking at unsuccessful voluntary there's 11 involuntary 26 so we know that there is a greater probability of observing an unsuccessful outcome with an involuntary referral then we would see with a voluntary referral so now let's take a look at logistic regression so I'm going to go to analyze regression and this case it's going to be binary logistic that's because we have a one dependent variable with two levels and this is what the logistic regression dialog looks like by default for dependent I'm going to move out come over and for covariates it's going to be gender and referral notice they're referred to as covariates and not predictor variables and when we put them in this covariance list box I want to click categorical up here and move them over to the categorical covariance list box notice the reference category is set to last on click continue here under save I'm going to under predicted values I'm going to click probabilities and group membership and I'll interpret these values they show up on the data editor and not in the output click continue under options I'm going to add classification plots the HL goodness of fit test and the confidence interval for exp B which here is actually beta so it's exp beta click continue and there's no changes under style so I'll click OK and here we have the output for the binary logistic regression we can see here that we have no missing cases and taking a look at the dependent variable encoding successful is coded to zero and unsuccessful is coded to one then for categorical variable coatings we can see we have the frequency and the parameter coding for each of the levels of the predictor variables there are 54 voluntary records 46 involuntary and 65 males and 35 females moving down the output we do want to see a statistically significant result here and we have that for variables in the equation the r-square can be found in the model summary we can see it's twenty seven point three percent twenty seven point three percent of the variance in the dependent variable can be explained by the predictor variables moving down to the HL test we want this to be a non statistically significant result and we have that point nine nine two is greater than point zero five and then we're going to move down to the variables in the equation and as we're looking at this table I want to take a look down a bit for predicted probability and you can see your predicted probability is of membership for unsuccessful so keep that in mind membership for unsuccessful so moving back to variables in the equation we can see we have gender one and referral one so this would represent gender one represent male and referral one would represent voluntary and you would find that in the data editor or in the variable view so you can see here for male and voluntary if I click a one see that male is zero and voluntary is zero as well now moving back to the output here for the p-value we want to make sure we have a statistically significant result for the predictor variables that we are going to interpret so we have a statistically significant finding here point zero zero eight is less than point zero five for gender and of course point zero zero one is less than point zero five for referral so we have statistically significant result for each of the predictor variables and the value that we're going to interpret here is going to be this exp beta and remembering here that we have the predicted probability is of membership for unsuccessful so again keeping that in mind as we begin this interpretation of exp beta the way we interpret this is we have gender male in this case for participant in that category the participant is going to be 4.25 1 times more likely to be unsuccessful than a female and notice here we're also provided the 95% confidence interval for exp beta the lower bound one point four six eight in the upper bound 12.30 8 then moving to referral remember this is going to be voluntary and notice the value for exp beta is less than 1 it's point two zero two so the way we would interpret this is to say that a participant in the voluntary referral group is us likely to be unsuccessful but an individual in the involuntary referral group now if you prefer to interpret this the other way to to interpret involuntary compared to voluntary instead of the voluntary value being here just go to a calculator and we're going to take one and divide it by the exp beta value here 0.2 0-2 and you can see that's four point nine five so it's just call that five so that tells us that if somebody's in the in voluntary category there are five times more likely to be unsuccessful than if they're in the voluntary category and again as I mentioned because of the way the output is configured it's a good idea in I believe most cases to run those chi-square tests so that you have an idea about the probabilities of group membership before beginning to interpret this table so taking a look at the data editor for the last part here we'll take a look at the two extra variables that I saved the predicted probability and the predicted group and because we have four combinations of gender and referral to dichotomous variables we're going to get four percents in this predicted probability variable you see the first one is about 32% so this is a 32% probability of being unsuccessful and that's female and involuntary together those two levels of these two independent variables combined gives you a 32% chance of being unsuccessful and because that's less than 50 the predicted group is going to be successful so this is the prediction that SPSS is making that the logistic regression is making based on this combination so a female who is involuntarily referred is predicted to be successful a male in the involuntary category has a 66% chance to be unsuccessful so the predicted group is unsuccessful a female that was voluntarily referred has just a point zero nine value here so that's nine percent just a nine percent chance of being categorized as unsuccessful so again the predicted group successful and for the last combination male and voluntary about a 28% chance point two eight for 28% chance of being unsuccessful so this combination is predicted to be successful I hope you found this video in conducting a binary logistic regression SPSS to be useful as always if you have any questions or concerns feel free to contact me and I'll be happy to assist you

Channel: Dr. Todd Grande

Views: 98,665

Rating: 4.8944101 out of 5

Keywords: SPSS, binary logistic regression, multinomial logistic regression, logistic regression, regression, dependent variable, independent variable, dichotomous, dichotomous variable, exp beta, exponentiated, beta, confidence interval, outcome variable, dichotomous outcome variable, variable, model fitting, odds, odds ratio, probability, group membership, predictor, hosmer-lemeshow, goodness of fit, covariate, chi-square, p value, significance, statistics, statistically significant, counseling, Grande

Id: iZoaXETWAN4

Length: 14min 44sec (884 seconds)

Published: Tue Jun 14 2016