Conducting an Ordinal Regression in SPSS with Assumption Testing

hello this is dr. Grande welcome to my video on conducting an ordinal regression in SPSS so before I start describing ordinal regression I want to briefly describe linear regression because there are many similarities between these two statistics so in a linear regression we're looking at the variance in a dependent variable also known as an outcome variable and we want to see how much variance independent or predictor variables explain in that outcome or dependent variable now in a linear regression the predictor variables and the outcome variable are all measured in either interval or ratio level of measurement which SPSS refers to as scale so with an ordinal regression we can conduct a regression using nominal or ranked data so nominal or ordinal levels of measurement and we can also include in the predictor variable a scale variable a variable measured at the interval or ratio level of measurement so I have fictitious data on this in this data view and we can see I have outcome pretest rubric and course so I'm going to switch to where you can see the labels so let's assume this data is gathered from some sort of training program and at the end of the training program raters rate the graduates of the program as either low medium or high in terms of the skills that they've developed as a result of that training and we have other variables that we've collected along the way we have a pretest which as you can see is scale we have a rubric which goes from zero two four so there's five different levels here unsatisfactory marginal satisfactory good and excellent and then we also have a course variable and this is just a zero or one so the zero represents that a course was not taken and a one represents that it was and let's just say this course would be believed to enhance a trainees ability to get a higher rating so it would be like an extra course added to the training program that you are considering including but you want to see how it's related to the outcome variable so let's set up the ordinal regression it would be analyze regression and then ordinal you can see linear is up here so we select ordinal and the dependent variable in this case is going to be the outcome so I'll move that over and you can see it's also referred to as status so you get to give it this outcome variable is the status of the trainee at the end you know their final rating and then we have a list box for factor and a list box for covariant so the factor list box would contain the ordinal or nominal variables and the co-vary it would contain the scale variables so first let's just run the factors or the ordinal or nominal level so we have the rubric zero through four and the course which would be zero or one and then over in options I'm going to leave this the same an output I'm going to add test of parallel lines that's the only change I'm going to make to the default display output settings and location I'm going to leave the same and scale I'm going to be the same so we'll hit okay to run the analysis so we can see in the analysis we have first the case processing summary which provides you with the n for the various score levels like in the rubric you have there was ten unsatisfactory scores sixteen marginal and so on and you can see we have no missing values all ninety scores were valid then moving down we have the model fitting information and we can see that we have a statistically significant result here Oh point zero zero zero so in this example for that for this table the moderate model fitting information we do want a statistically significant result we want less than point zero five so this is a good finding in terms of how well does our model fit the data then moving down to goodness of fit now for this table we would like to fail to reject the null hypothesis so we're looking for significance value greater than point zero five and as you can see we have that for the first p-value but not for the second so this is a mixed result here the next table looking at the R square I'll interpret the middle value here 0.56 one tells us that our model explains fifty six point one percent of the variance in the dependent variable and then before we get to parameter as estimates I want to take a look at test of parallel lines this tests the assumption of proportional odds and we want this to be greater than or equal to point zero five and in this case we can see that we violated this assumption we have a statistically significant result of 0.0 is caution when interpreting the output and then you can see we have threshold for outcome these are not values that we would typically interpret in this output of more interest would be location so we have the estimates for location for rubric and course and of course we look at significance so we can see that for all the rubric scores except for rubric equals to or should say for rubric equals one through three we have a statistically significant result except for the value two and for the course equaling zero meaning the participant did not take the course we have a statistically significant result you can see that the last value the last level is made into the reference variable that's why rubric equals four is zero and course equals 1 is zero so as we compare these estimates to the reference level we need to recognize here what these values mean and if we have a value like this that's lower than zero which this is it's negative one point five eight one that indicates that lower cumulative scores are more likely and you can see that's the case for rubric equalling one all the way through three we would expect lower scores here as lower kumo scores are more likely with a negative value and the same thing for the course so of course equals one would they taken the course as the reference variable and the course equals zero negative three point two six nine a lower kumo score would be more likely in the situation for this level and of course if these values were positive if any of these values were positive for that level we would assume higher cumulative scores we're more likely I also want to run this ordinal regression including a scale variable so I'm going to go back to analyze regression and ordinal and I'm leave everything the same except now I'm going to add the pretest which is in scale the scale level measurement I'm going to add that as a covariant and hit OK and we can see we still have a good model fit and we have here a goodness of fit which is much greater than point zero five and moving down here to the test of parallel lines now we fail to reject the null hypothesis which is the result that we would want we want this value be greater than point zero five but I wanted to show you how pretest is now in location and you see it doesn't have any levels it's a scale and notice how the estimates are affected a little bit they're different by a little bit but the pretest estimate here is just around zero and we it's not statistically significant it's point nine seven three so I want to show you what the parameter estimates look like when you do include a variable that was recorded at an interval or ratio level of measurement I hope you found this video on conducting an ordinal regression to be helpful 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: 112,741

Rating: 4.8246756 out of 5

Keywords: SPSS, ordinal regression, linear regression, assumptions, tests, parallel lines, null hypothesis, goodness of fit, model fit, outcome, predictor, variable, data, analysis, estimate, nominal, ordinal, interval, ratio, scale, counseling, Grande, Regression Analysis, Ordinal Data, Statistics (Field Of Study)

Id: ioNr9o8v5o0

Length: 10min 50sec (650 seconds)

Published: Fri Aug 21 2015