tutorial 016 Generalized linear model

welcome to the sdss tutorial video on the generalized linear model the generalized linear model is an expansion of the general linear model which is often referred to as a regression it uses different types of outcome variables in this video we'll be covering logistic regression which is a regression under the generalized linear model using a binary or yet no type of outcome variables we'll also discuss options for how to run Poisson distribution and other types of options within the generalized linear models for this video we'll be using the therapy data set used in previous videos however I've created a new variable the final one listed here called depression at time to high low for this variable as you can see under the values I've created a new variable such that all the scores for depression at time 2 that are below the mean we're now coded as zero or low depression group I've next coded all the scores that were above the mean under a high depression group scored as one will be using this binary outcome variable for our logistic regression to run a logistic regression you can do it in multiple ways but we'll be covering today is the generalized linear model the first page you see under these tabs at the top of the screen within the generalized linear model is the type of model will be running here if we specify a linear scale it's just like running a typical regression we can also run generalized linear models with ordinal responses that is responses that are categorical but in certain orders count variable that is the number of behaviors of person shows in the study or in this case a binary variable a binary logistic variable for logistic regression we make need to indicate which is our outcome variable our dependent variable as I said before this depression at time 2 scored high and low and for any binary variable SPSS is going to ask you to specify a reference category you can choose either the highest value for the one in this case because we coded 01 or the lowest value which poor should be zero the only thing this will change in the model is the sign the positive negative of the coefficients within the model other than that sand math is the same after we set fire dependent variable as well as the reference category we need to add predictors to role models unlike the regression menus the generalized linear model splits predictors into two types it includes factors or categorical predictors and covariance or continuous predictors for the sake of this example allies one of each let's add participants X as our factor as our categorical predictor and optimism is our covariance now we also need to include which variables will be included in the model we need to specify that both female and optimism will be included in for a model we can also choose under this building terms whether we want only the main effects included in the model or whether we also want to include the interaction between these two attorneys for the sake of this example I'll choose the interaction now the other tab seen above estimation statistics estimated marginal mean save and export provide a great deal of more sophisticated options for how to run specific models within maximum likelihood for how to output specific types of statistics but for most users what I've shown so far within these first four tabs will be sufficient and we'll run this analysis as with most output in SPSS it first reminds you of what you ran our dependent variables depression at time to with high and low we chose a binomial distribution in the logit link function this is for logistic regression again your instructor will discuss other link functions and probability distributions such as Poisson distributions in classes we're reminded of how many cases are included we have 150 people and under our categorical information it includes how many of our dependent variables were listed as high and low we had 77 people who are in high depression and 73 who are low in depression next we see the descriptive statistics for our continuous variable of optimism would eat this table we see a table for goodness of fit this is the chi-square goodness of fit values that we see used in the omnibus test below for the likelihood chi-square ratio this omnibus test is whether we find out whether the overall model is statistically different from 0 below the test of the overall model we see specific tests for each of the variables we can see here that female or our sex variable was not a significant predictor of depression categories as higher low our optimism variable with a marginally significant predictor with t less than Clinton 10 but not significant at p less than 2005 we also get parameter estimates it looks much like a regression table for our individual predictors now we've completed a logistic regression and i'll show you briefly how to run a poisson regression although we won't run the actual output will return to analyze and the generalized linear model for our type of model instead of specifying a binary logistic we'll choose a plus on distribution

Channel: Dustin Thoman

Views: 8,899

Rating: 4.5454545 out of 5

Keywords:

Id: vp3iscJe-_s

Length: 5min 41sec (341 seconds)

Published: Thu Feb 09 2017