SPSS Tutorials: Binary Logistic Regression

welcome to the LSC methodology Institute SPSS tutorial series sponsored by the LSC Annual Fund in this video we'll give you a brief introduction to binary logistic regression for this example of binary logistic regression we'll be using this variable here petition as a response variable where we asked respondents whether or not they've ever signed a petition we've coded 0 for no and 1 for yes to that question 4 explanatory variables we're using this variable here vote where we asked respondents whether or not they intend to vote with 0 for no and 1 for yes the number of voluntary groups in which the respondent is involved the respondents gender with 0 for female and 1 for male and finally the respondents age in years so to carry out a binary logistic regression in SPSS we click analyze regression and binary logistic so first we want to put in our response variable which is petition so we click on it and we click on this arrow and we see that it appears in the box under the word dependent to get our explanatory variables we click on save vote and then click on this arrow here and then this is the question about voluntary groups we click on it and then again click on the same arrow click on the respondents gender click again on the arrow and finally the respondents age and then click on the arrow in this case we've coded all of our categorical variables as 0 & 1 such as gender and vote but if we had a different coding say if we had a categorical variable with more than two categories what we can do for this type of analysis is click on this button here that says categorical and then say we choose vote and then click on this arrow and then what we can do is we can indicate which category we have as our reference category and then SPSS will do the rest but we won't do that this example the other thing we want to do before we actually click on OK is click on options and we also want to display the confidence intervals so we click on this checkbox here and we see that the default option is 95 so we're going to get the 95% confidence intervals as well so we click on continue we click on OK and then we let SPSS do its thing and here's what we see in the output window you'll notice that there are many tables containing various pieces of information and I won't go through each one of them but I would just point out that these subheadings here are useful when you're testing blocks of models against each other so in this very simple example where we've just run a single model to find the parameter estimates for the model that we've just run we need to select the very last table in the list and here you'll see that we have a row for each of the explanatory variables in our model with information about their constant or the intercept contained in the last row of the table so let's take a very quick look at the relationships between the explanatory variables and our response variable remember voters of binary variable coded one if you intend to vote in the next election and 0 if not now we can express the relationship between the intention to vote and signing a petition in two different ways we can say that the intention to vote increases the logit or the estimated log odds of signing a petition by 1.59 to unit and we can also express this same relationship in terms of odds ratio so if we take the exponential of 1.5 92 we'll obtain four point nine one five so we can say that those who intend to vote in the next election are nearly five times as likely to sign a petition as those who don't intend to vote in their next election and in the last two columns we have the lower and upper limits of the 95% confidence interval which is two times of a lower level and about 11 times at the upper level and then those middle columns in the table give us information about the significance test for that estimated coefficient so here we're testing the null hypothesis that in the population there's no difference in the logarithm of the odds of signing a petition for voters compared to non voters in other words this coefficient is zero or this odds ratio is one and here we have the standard error for the estimate of the coefficients and we have the test statistic which is a Wald pesto's ethic in this case on one degree of freedom and a p-value which is very small so at any conventional significance level we could say we could reject that null hypothesis and infer that controlling for all of the other variables in our model there is a relationship between our intention to vote and your likelihood of signing a petition looking to the next row of the output this is information about our voluntary groups the number of voluntary groups that you belong to and this coefficient indicates that for every unit increase in that variable in other words for every extra voluntary group that a person belongs to the model estimates that they are load yet that is the log odds of their signing repetition increases by 0.47 for units by nearly half a unit and that translates into multiplying the odds of signing a petition by 1.6 so you take the exponential of 0.47 for you'll obtain one point six zero six so in other words controlling for all of the other variables in the model for every extra voluntary group that a person belongs to their odds of signing a petition increased by about 60% and again if you like you can report the confidence in front a 95% confidence interval for that estimate and again we can test the null hypothesis that in the population controlling for all of the other variables in the model there is no relationship between number of voluntary groups phone sign your petition which once again would of course imply that this coefficient is zero or this er odds ratio is one and here's our test statistic Wald sadistic and our p-value which is a here point zero 1 1 which seems to me pretty strong convincing evidence against the null hypothesis so you could if you are using significance levels say that you reject the null hypothesis at the 5% level but of course you couldn't reject it not quite at the 1% level it will just be borderline now skimming quickly to the p-values for the other two expansion variables in the model you can quickly see that they're fairly high higher than any conventional significance level that we might use for inferring a relationship between those two variables and the response variable and you can see correspondingly that the estimates of their of their coefficients on the logit scale are very close to zero which translate into odds ratios very close to one so the message seems to be that controlling for the other variables in the model there isn't any significant relationship between gender and petition signing nor between age and petition sign so that was just a very quick introduction to binary logistic regression running a very simple model with two binary explanatory variables and do continuous explanatory variables and taking a quick look at the main results from SPSS and we'll leave it there for now

Channel: Department of Methodology LSE

Views: 328,076

Rating: 4.7176471 out of 5

Keywords: SPSS, SPSS Tutorial, software tutorial, data, Binary Logistic Regression, data analysis, SPSS analysis, Methodology Institute, statistical software

Id: Ak_t86zm_sQ

Length: 7min 40sec (460 seconds)

Published: Tue Jul 03 2012