Dear list,
Firstly, thank you so much for immensely helpful replies about my question on homoscedasticity assumption in the context of multiple regression. I have used the NCV test available in car package for r, and the Breusch and pagan test to assess the assumption of homoscedasticity. Both tests are highly significant, chi-square values above 10, at 1 degree of freedom. I have a dependent variable which is a sum of six likert items. The response scale was from 1 (never) to 5(always). I’ve multiple independent variables, all have been measured using likert items, and scores on those items have been summed to form scales. The sample size is 1250. I am reading discovering statistics using SPSs by Field and R in action data analysis and graphics with R. Both books suggest that when homoscedasticity is violated, the standard errors may not be correct, and tests of significance may not be optimum. In such a situation bootstrapping is suggested. I have performed bootstrap both using SPSS and R. My question may be the dumbest question ever asked on this list, so apologies in advance. How to interpret results of bootstrap in SPSS and how to report those results. In the output of SPSS, I have two tables for coefficients. First is the standard table of coefficients that spss provides in normal regression output, and the second has bootstrap confidence intervals, and standard errors. What I’m unable to work out is that firstly, can I use the T values in the coefficients table, as there is no T value at least that is what I can understand, in the bootstrap table. secondly, in spss output of bootstrap, the model summary, anova, and coefficients table appear twice. I assume that the first set of tables is before bootstrap, and second set of tables are a part of bootstrap regression. However, the F value, in the anova table, and the adjusted r-square are completely unchanged. Is this how it should be? Lastly, I’m envisaging the regression results table that I’ll construct, I believe that the unstandardized coefficients, standard errors, will come from the bootstrap coefficient table, where to take the T value from? Standard coefficient table? The coefficients table also has sig column and the bootstrap results table also has sig column, which sig values to report? Again, sorry for such a basic question. Regards, Faiz. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In my work, I have found the bootstrapping is effective for dealing with non-normality, but not for heteroscedasticity... To deal with that, I think you need to look at the rlm() function in the MASS package or the use of "sandwich estimation" which requires the car and lmtest packages. All done in R.
wbw William B. Ware, Ph.D. McMichael Term Professor of Education, 2011-2013 Educational Psychology, Measurement, and Evaluation CB #3500 - 118 Peabody Hall University of North Carolina at Chapel Hill Chapel Hill, NC 27599-3500 Office: (919)-962-2511 Fax: (919)-962-1533 Office: 118 Peabody Hall EMAIL: [hidden email] Adjunct Professor, School of Social Work Academy of Distinguished Teaching Scholars at UNC-Chapel Hill -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of faiz rasool Sent: Tuesday, February 27, 2018 8:20 AM To: [hidden email] Subject: how to interpret bootstrap results of multiple regression Dear list, Firstly, thank you so much for immensely helpful replies about my question on homoscedasticity assumption in the context of multiple regression. I have used the NCV test available in car package for r, and the Breusch and pagan test to assess the assumption of homoscedasticity. Both tests are highly significant, chi-square values above 10, at 1 degree of freedom. I have a dependent variable which is a sum of six likert items. The response scale was from 1 (never) to 5(always). I’ve multiple independent variables, all have been measured using likert items, and scores on those items have been summed to form scales. The sample size is 1250. I am reading discovering statistics using SPSs by Field and R in action data analysis and graphics with R. Both books suggest that when homoscedasticity is violated, the standard errors may not be correct, and tests of significance may not be optimum. In such a situation bootstrapping is suggested. I have performed bootstrap both using SPSS and R. My question may be the dumbest question ever asked on this list, so apologies in advance. How to interpret results of bootstrap in SPSS and how to report those results. In the output of SPSS, I have two tables for coefficients. First is the standard table of coefficients that spss provides in normal regression output, and the second has bootstrap confidence intervals, and standard errors. What I’m unable to work out is that firstly, can I use the T values in the coefficients table, as there is no T value at least that is what I can understand, in the bootstrap table. secondly, in spss output of bootstrap, the model summary, anova, and coefficients table appear twice. I assume that the first set of tables is before bootstrap, and second set of tables are a part of bootstrap regression. However, the F value, in the anova table, and the adjusted r-square are completely unchanged. Is this how it should be? Lastly, I’m envisaging the regression results table that I’ll construct, I believe that the unstandardized coefficients, standard errors, will come from the bootstrap coefficient table, where to take the T value from? Standard coefficient table? The coefficients table also has sig column and the bootstrap results table also has sig column, which sig values to report? Again, sorry for such a basic question. Regards, Faiz. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
I should point out that the GENLIN procedure (Analyze > Generalized Linear Models > Generalized Linear Models) allows you to run a linear model estimating the covariance model robustly and thereby get good standard errors/significance levels. On Tue, Feb 27, 2018 at 8:27 AM, Ware, William B <[hidden email]> wrote: In my work, I have found the bootstrapping is effective for dealing with non-normality, but not for heteroscedasticity... To deal with that, I think you need to look at the rlm() function in the MASS package or the use of "sandwich estimation" which requires the car and lmtest packages. All done in R. |
Free forum by Nabble | Edit this page |