Interpreting Odds Ratio for Multinomial Logistic Regression using SPSS - Nominal and Scale Variables

hello this is dr. Grande welcome to my video on interpreting odds ratio in multinomial logistic regression using SPSS as always if you find this video useful please like it and subscribe to my channel I certainly appreciate it I have in the SPSS data editor fictitious data and I'm going to be focusing on three variables here age type and modality so a multinomial logistic regression allows us to use independent variables and predict group membership in a dependent variable and the levels of a dependent variable in the case of a multinomial logistic regression we can have three or more levels if we only had two levels to the dependent variable we would use a binary logistic regression so in this example for modality the dependent variable I have three levels 0 1 and 2 that corresponds to CBT psychodynamic and Gestalt and you can see I'm moving back and forth between the string and the number using this a one button up top then we have the two independent variables and one is continuous that's age and the other is categorical and this is type this there are three types individual group and family again 0 1 and 2 so we have three levels of the independent variable type three levels of the dependent variable modality and for the independent variable age this is continuous so for this example let's assume that we want to predict group membership in the dependent variable in one of these three types of therapies CBT psychodynamic or Gestalt and we're going to be using age to try to predict group membership and type and for this example let's just assume that every participant has only been exposed to one type of therapy individual group or family and that we believe that because they've only had this one type of therapy and because their age these two variables that's going to influence what they're going to select in terms of modality so these participants will be selecting a modality CBT psychodynamic or Gestalt and we believe that their age and the type of counseling that they've been exposed to will allow us to predict what level they select of the dependent variable what odhh ality they select so now I'm going to perform the multinomial logistic regression and we'll take a look at the odds ratio so I'll go to analyze and regression and then multinomial logistic here's the dialogue for multinomial logistic regression for the dependent variable we want to use modality these participants will be selecting one of these three levels and modality so I'm going to move that over and for the reference category I'm going to leave it set to the default which is last you can click this reference category button and select first category or set a custom category I'm just going to leave it set to last then for type type is nominal its categorical it will move over to factor this factor list box age is continuous so age is going to move over to the covariant list box so both a age and type are independent variables but they're treated differently because of the level of measurement types categorical so it goes and factor age is continuous so it goes under covariant so if you have an ordinal variable and ordinal variable in a multinomial logistic regression needs to be treated as nominal or as continuous in this example of course I don't have any ordinal variables so the way I have it set up here is fine so I'm going to perform the multinomial logistic regression here by clicking ok so here we have the output for the logistic regression and I'm going to be focusing on this last table parameter estimates so you see we have two major rows here for modality we have CBT and psychodynamic notice Gestalt is not in this list and if you look down here the reference category is Gestalt so you have these two categories these two levels of the dependent variable modality CBT and psychodynamic and then you have a reference category and that's Gestalt looking here at the age and the type you see we have age for both CBT and psychodynamic and the three types for both levels of the dependent variable type 0 is individual type 1 group and type 2 family so first we'll interpret the odds ratio for the continuous variable in this case age and of interest here will be exp beta that's the odds ratio so we looking at modality and these independent variables over here to the left and to the right exp beta so here for age and modality CBT we can see the exp beta value is one point zero two so what this means is that for every one year increase we have in age this multiplies the odds of selecting CBT rather than Gestalt by one point zero two it increases the odds by two percent if we were to look at a ten year change in age that would be two percent times 10 or 20 percent that would increase the odds by 20 percent again of selecting CBT rather than Gestalt Gestalt is the reference category all this of course assumes that type is held constant so we interpret exp beta for age as I've just described when type is held constant similarly looking at age for the psychodynamic level we have exp beta value of one point zero eight three so that represents an eight point three percent increase meaning for every one year added to age the odds of the participant selecting psychodynamic rather than Gestalt increase by eight point three percent so now let's take a look at the categorical variable and that's type so again type zero is individual type one is group and type two is family so we'll start here with type zero that's individual so I'm going to move over here to the X P beta and you can see the value there is two point six eight three so this tells us that if we hold age constant the odds of a participant who only received individual counseling this type 0 type equals zero selecting CBT rather than Gestalt is two point six eight three times higher than the odds for somebody in type two somebody who only received Family Therapy so we're comparing here individual and family now if we're looking at group that's type 1 this is one point eight one three so comparing somebody in the with the group type to the family type the odds of somebody in the group type selecting CBT rather than guessed all is one point eight one three times greater than somebody in the family type so again both the individual type and the group type here are being compared to the family type and again if we want to convert this to percent the two point six eight three would be one hundred and sixty eight point three percent the one point one eight three would be eighty one point three percent so moving down here to type zero individual in psychodynamic see this value is below one point three five seven so we take one and subtract this value it gives this point six four three or sixty four point three percent so this tells us that a participant who's been exposed to individual counseling has 0.357 times the odds of selecting psychodynamic as opposed to Gestalt or sixty four point three percent less so moving back up to CBT we know how to compare type zero individual and type one group two type two which is family but can we with this table compare type zero individual to type one group and answer is yes the odds of somebody who's been exposed to individual counseling selecting CBT rather than Gestalt are 1.47 times higher then for somebody in the group type so how did I arrive that value one point four seven nine so I'm going to calculator over here so if we look at the exp beta for type 0 individual that's two point six eight three and then we're going to divide it by the exp beta for type one one point eight one three that gives us one point four seven nine so what if we want to see the odds of somebody selecting CBT rather than psychodynamic right so not using the reference category Gestalt but just looking at CBT and psychodynamic let's use the example of type one group compared to family type two so we have for type 1 and CBT exp Beit of one point one eight three and for type one psychodynamic exp beta of 0.295 so moving back to the calculator this would be one point eight one three divided by 0.295 it gives us six point one four five so again holding age constant the odds for somebody who had the group counseling selecting CBT rather than psychodynamic our six point one four five times the odds for somebody in the family type so again in this example we are comparing the group type to the family type I hope you found this video on interpreting odds ratio and multinomial logistic regression to be useful 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: 75,701

Rating: 4.9198217 out of 5

Keywords: SPSS, multinomial logistic regression, logistic regression, binary logistic regression, odds ratio, odds, ratio, exp beta, exp, beta, parameter estimates, reference category, reference, prediction, group membership, dependent variable, independent variable, predictor variable, predictor, continuous variable, continuous, scale variable, scale, ordinal variable, ordinal, outcome variable, variable, model fitting, probability, statistics, analysis, counseling, Grande

Id: ov8aZmD9i8E

Length: 13min 46sec (826 seconds)

Published: Wed Dec 14 2016