How to Use SPSS-Hierarchical Multiple Regression

in this video I will demonstrate how to perform hierarchical multiple regression in this technique the overall goal is to use several independent or predictor variables to predict a single outcome but in this case in hierarchical regression also known as sequential regression the independent or predicted variables predictor variables are entered into the equation in the order specified by the researcher based upon certain theoretical or preferential grounds so variables or sets of variables are entered in steps or what's known as blocks with each independent variable being assessed in terms of what it adds to the prediction of the outcome or the dependent variable after the previous variables the variable is entered in first are controlled for so this is a way to control for the effects of variables that you think might be confounding or covariant types of variables so that we can account for that and again develop a better prediction model for the outcome so for example if you wanted to know how well a grade point average in undergraduate or standardized test scores would predict someone having a certain score on a licensure or certification exam or if you wanted to determine how well flexibility and muscle strength predicted vertical jump or running speed but we also wanted to control for the effect of age or weight or some other potential confounding variable we could enter our count founding variables into the first block and then our actual predictor variables we want to use to predict the outcome in the second block and then once all the sets of variables are entered the overall model is assessed in terms of its ability to predict the outcome or the dependent measure and so the relative contribution of each block of variables is also assessed and we can determine very precisely how much influence particular particular variables may be having now the assumptions for hierarchical multiple regression are the same as they are for standard multiple regression we still have sample size issues we need to monitor we have multicollinearity we need to track and monitor we have to look for outliers and then we also have to look for normality and linearity of the outcome variables with the predictor variables so I'm not going to go into a lot of depth with examining the assumptions in this example you can see a description of that in standard multiple regression video and the procedures are exactly the same as far as evaluating assumptions whether we're doing standard multiple regression or hierarchical medical multiple regression so I'm going to focus more in the interpretation of the model that we produce when we do hierarchical multiple regression so the example we're going to do for hierarchical multiple regression is we're going to use some data in which we're trying to predict someone's perceived stress level as measured by this stress index with a higher score being a higher level of perceived stress and we're going to use a predictor variable that measures someone's perceived mastery of external events also someone's perceived ability to monitor and control internal moods such as physiological responses mood responses and that sort of thing and then we're also going to control for age and we're going to control for an index of social interaction how well they feel they interact socially so we're going to use their perception of their internal controls and their perception of their external controls to predict stress while controlling for a measure of social interaction and also controlling for age so a research question we could come up with from this is if we wanted to control for the possible effect of age and social interaction is our Purdue our predictor variables are they still able to predict a significant amount of variance in perceived stress so to address this question we are going to do hierarchical multiple regression so we'll be entering our variables into in steps or blocks in a predetermined order so we're not letting the SPSS decide what order to enter them and we're deciding so in our first block we will force age and social interaction into the model okay so this has an effect of statistically controlling for these variables and then the second step or the second block we will force in our two predictor or independent variables into the model just as we did in the previous example of standard multiple regression so the difference this time is that the possible effect of age and social interaction has been removed or controlled for and then we can see whether our block of predictor variables are still able to explain some of the remaining variants in our dependent variable so that's going to be our overall process so to actually do this in SPSS we need to go to the analyze menu go to regression and choose linear so we're going to take our outcome variable or a dependent variable which is perceived stress and move that into the dependent variable box okay we're going to move our control the the variables you wish to control into the independent box so we have social interaction in age so this will be the first block of variables that we're entering in the analysis so block of one or block number one alright then we click on the button that says next and then we enter in the variables that will be our predictor variables or in the second block so that's our mastery variable in our pcoi SS variable okay so we're entering in the variables in two blocks the first block are the variables that we want to control and the second block of the variables that we want to use is our predictor or independent variables okay now in the method box make sure we have enter checked okay which is the default again this we're forcing them in a specific order okay then we then click on the statistics button here okay we want to make sure we have checked estimates model fit r-squared change descriptives part and partial correlations and then collinearity Diagnostics okay and we go ahead and click continue now under the options button in the missing values section click on exclude cases pairwise so again if we've got a subject that's missing data on one or more of the variables that subject will be excluded from the analysis let me click on okay orbs are continue and now we click on the plots button okay and we're first going to click on the variable that's titled star s I'm sorry zr e SI d which is this one here we move it into the y box and then we find the variable titled star espy re d and move that into the x box okay where it says standardized residual plots we're going to tick the normal probability plot box make sure that's checked then click continue now we're going to click the Save button and we're going to click on look under distances and click on the first option maha la la la I can never save this Mahalanobis and then Cooke's okay then we click continue and then we click ok so as our output generates again our first step in this process is always going to be evaluating the assumptions that we talked about for this analysis so I'm kind of skipping that step we're going to assume we've done that and again you can get an in-depth description of how to do that in the video for standard multiple regression so I'm going to go ahead and go into the assumption that we've met all of the assumptions to do this hierarchical multiple regression so our first step then is to evaluate the model to see if the model actually is able to predict the outcome we want to predict so we go to the model summary table first alright and we check the r-squared values so as we see the variables in block 1 these are the variables that we want to control for they account for about 5.7 percent of the variance in the outcome so we take that point oh five seven number multiply it by a hundred and this is saying to us that the variables that we entered into model number one account for about 5.7 percent of any variability in our outcome now after we enter the block two variables okay these are actual predictor variables now we can see that the model as a whole now explains about forty seven percent of the variability in perceived stress so it's important to note that this second R square value includes all the variables from both blocks not just those in the in the second block so now we've incorporated all four variables and so we're controlling four variables in the first two block or in the first block and now we're seeing what effect all the variables have together after that control so to find out how much of this overall variance is explained by our predictor variables the variables were interested in mastery and P oh I see SS after the effects of age and social interaction have been removed we can look in the column labeled R square change which is this variable right here okay so in the output presented you'll see that on the line marked model to the bottom here that the R squared change value is point four one seven so this means that our our predictor variables are independent variables we were interested in explained an additional 42 percent of the variance in our outcome even when the effects of age and social interaction have been statistically controlled for so it adds a significant amount of variance prediction so this is and if we look over at the si gf change we can see that this is statistically significant contribution so this si gf change score is less than 0.05 okay and so this indicates to us that the addition of these two predictor variables now has a statistically significant contribution to predicting the outcome now if we look at the ANOVA table just below it and we look at model number two we can look at the SI g value and this again tells us how the model is a whole is able to predict including all four variables we can see that this model is a statistically significant predictor of our outcome again less than the point O 5 and less than 0.01 level so this model is a statistically significant predictor of perceived stress so what we've done again is we've controlled for two confounding variables and then we've added in our two predictor variables and now this model while controlling for the confounding and using the predictor variables is a statistically significant predictor of the outcome of interest in this case perceived stress okay so our second step then is evaluating each of the independent variables so to find out how well each of the variables contributes in the final model we need to look in the coefficients table okay so we find the coefficients table click on that and we're going to focus here on the model two row okay so this summarizes the results with all the variables entered in the equation so as we look at the SI g column over here there are only two variables that make a unique statistically significant contribution in other words the SI g values are less than 0.05 and you can see that those are two predictor variables are two independent variables make a statistically significant contribution to the model the social interaction and the age variables do not make a statistically significant contribution so if we look at our standardized coefficients column beta and we look at each of the variables we can see which of our two predictor variables has the makes the most contribution okay the the largest unique contribution and that is the mastery score and the PC oh is s score is a little bit less but these two variables both make a statistically significant unique contribution to the model now neither age nor nor social interaction make a unique contribution now remember these beta values represent the unique these beta values here represent the unique contribution of the variable when the overlapping effects of all the other variables have been statistically removed so in different equations with different sets of independent variables or maybe if we added some additional predictor variables or additional confounding variables that we control for and model one these these values would change if we had a different sample size these values would change potentially so we have to bear in mind that that this identification of the contribution of these variables is specific to this this situation this sample in this collection of variables now as far as presenting results or making conclusion there's a couple different ways we can do this but at minimum we wanted what we want to do is report or indicate what type of analysis was performed we did a hierarchical multiple regression we want to report the standardized beta values we can also report the beta coefficients the B coefficients if we want to use a actual multiple regression equation along with their standard errors we can report the ANOVA results we can report the r-squared values and we want to do that at each step so we're on report those descriptives that with it the first step model one as well as that model - so to summarize we can use hierarchical multiple regression when we want to again make a prediction of a single quantitative or or continuous outcome using multiple predictor or independent variables while controlling for potential confounding variables by doing a a block pattern of entering the variables into the model so the first block will always be the variables that we want to control for and then the second block would be the variables that we want to use as the actual independent variables our actual predictor variables so just like with standard multiple regression we still have to check for assumptions so that's always for step in this process but hopefully you were able to learn something from this presentation and hopefully you have good luck using this technique in your future research

Channel: TheRMUoHP Biostatistics Resource Channel

Views: 158,217

Rating: 4.884131 out of 5

Keywords: statistics

Id: xgA8qY63dX0

Length: 18min 0sec (1080 seconds)

Published: Thu May 09 2013