Correspondence
Comment on Article Entitled: “PLAY Project Home Consultation Intervention Program for Young Children with Autism Spectrum Disorders: A Randomized Controlled Trial” To the Editor: We recently saw the article published in JDBP entitled “PLAY Project Home Consultation intervention program for young children with autism spectrum disorders: A randomized controlled trial” (volume 35, issue 8) by Solomon et al.1 The reason why we ran into this article was that we were looking for articles citing our own publication in Multivariate Behavioral Research, entitled “Analysis of variance of multiply imputed data” (volume 49, issue 1).2 Although we were happy with the citation, we were a little bit less happy with the way it was cited. On page 480, it is stated that: “Although multiple imputation procedures are promising,3 the field has not advanced to provide methods for addressing repeated measures analysis of variance (ANOVA) analysis,2 as used in this article.” This phrase suggests that we stated that for multiple imputation no solution exists for repeated measures ANOVA, whereas the goal of the article was to actually provide the reader the complete procedure for multiple imputation and (repeated measures) ANOVA. We even provided the tools to carry it out in several software packages, including an SPSS syntax file4 to do it in SPSS 19.5 What we did say was that the procedure for repeated measures ANOVA was far from trivial so that its standard implementation in SPSS would be imperative (rather than using the additional tools that we provided) for this procedure to become common
Vol. 36, No. 3, April 2015
practice. But that is not what the authors quoted us on. Through this letter, we hope to get rid of the possible misunderstanding that we stated that there are no tools for doing repeated measures ANOVA on multiply imputed data sets. Joost R. van Ginkel, PhD Pieter M. Kroonenberg, PhD Department of Child and Family Studies Leiden University Leiden, the Netherlands REFERENCES 1. Solomon R, Van Egeren LA, Mahoney G, et al. PLAY Project Home Consultation intervention program for young children with autism spectrum disorders: a randomized controlled trial. J Dev Behav Pediatr. 2014;35:475–485. 2. Van Ginkel JR, Kroonenberg PM. Analysis of variance of multiply imputed data. Multivariate Behav Res. 2014;49:78–91. 3. Schafer JL. Multiple imputation: a primer. Stat Methods Med Res. 1999;8:3–15. 4. Van Ginkel JR. MI-MUL2.SPS [computer code]. Available at: http://www.socialsciences.leiden. edu/educationandchildstudies/childandfamily studies/organisation/staffcfs/van-ginkel.html. Accessed February 5, 2015. 5. SPSS Inc. SPSS 19.0 for Windows [computer software]. Armonk, NY: IBM Corp; 2010.
Authors’ Response: In our article, we indicated that “the field has not advanced to provide methods for addressing repeated measures ANOVA analysis” and cited van Ginkel and Kroonenberg.1 It should be clarified that this refers to the assertion by van Ginkel and Kroonberg1 that “the implementation of the combination rules needed for ANOVA in statistical software packages has either been quite limited or very well hidden.” It should also be clarified that van Ginkel and Kroonberg1 have proposed methods to conduct the analyses using additional macros and explanations not integrated into major
statistical software packages and argue for the inclusion in future upgrades. As van Ginkel and Kroonenberg1 state: “To date, the implementation of the combination rules needed for analysis of variance (ANOVA) in statistical software packages has either been quite limited or very well hidden. SPSS 19.0 does not provide any pooled results for F tests for multiply imputed data. Using an SPSS macro (Van Ginkel), the procedure may be carried out in SPSS but this still requires substantial manual preprocessing. For repeated measures ANOVA, the procedure is even more complicated. Although SAS 9.3 and Stata 10.0 have a procedure for carrying out the combination rules for ANOVA in multiple imputation, the manual provides no clear practical guidance in how to use them in the specific context of ANOVA. The procedure outlined in this article should also be of assistance to any SAS or Stata user who intends to use multiple imputation in this context. In the appendix, it is explained how exactly the procedures can be carried out in SAS 9.3, Stata 10.0, and SPSS 19.0 using the macro of Van Ginkel. Considering the involved task needed to actually carry out the described procedure, its implementation in future releases of SPSS and other software packages, as well as a clear guidance with examples in the SAS and Stata manuals, seems desirable if its use is to become commonplace.” Richard Solomon, MD* Laurie Van Egeren, PhD† Melisa Quon Huber, PhD† *Ann Arbor Center for Developmental and Behavioral Pediatrics Ann Arbor, Michigan †Michigan State University E. Lansing, Michigan REFERENCE 1. van Ginkel JR, Kroonenberg PM. Analysis of variance of multiply imputed data. Multivariate Behav Res. 2014;49:78–91.
www.jdbp.org | 225
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Comment on Article Entitled: “PLAY Project Home Consultation Intervention Program for Young Children with Autism Spectrum Disorders: A Randomized Controlled Trial” To the Editor: We recently saw the article published in JDBP entitled “PLAY Project Home Consultation intervention program for young children with autism spectrum disorders: A randomized controlled trial” (volume 35, issue 8) by Solomon et al.1 The reason why we ran into this article was that we were looking for articles citing our own publication in Multivariate Behavioral Research, entitled “Analysis of variance of multiply imputed data” (volume 49, issue 1).2 Although we were happy with the citation, we were a little bit less happy with the way it was cited. On page 480, it is stated that: “Although multiple imputation procedures are promising,3 the field has not advanced to provide methods for addressing repeated measures analysis of variance (ANOVA) analysis,2 as used in this article.” This phrase suggests that we stated that for multiple imputation no solution exists for repeated measures ANOVA, whereas the goal of the article was to actually provide the reader the complete procedure for multiple imputation and (repeated measures) ANOVA. We even provided the tools to carry it out in several software packages, including an SPSS syntax file4 to do it in SPSS 19.5 What we did say was that the procedure for repeated measures ANOVA was far from trivial so that its standard implementation in SPSS would be imperative (rather than using the additional tools that we provided) for this procedure to become common
Vol. 36, No. 3, April 2015
practice. But that is not what the authors quoted us on. Through this letter, we hope to get rid of the possible misunderstanding that we stated that there are no tools for doing repeated measures ANOVA on multiply imputed data sets. Joost R. van Ginkel, PhD Pieter M. Kroonenberg, PhD Department of Child and Family Studies Leiden University Leiden, the Netherlands REFERENCES 1. Solomon R, Van Egeren LA, Mahoney G, et al. PLAY Project Home Consultation intervention program for young children with autism spectrum disorders: a randomized controlled trial. J Dev Behav Pediatr. 2014;35:475–485. 2. Van Ginkel JR, Kroonenberg PM. Analysis of variance of multiply imputed data. Multivariate Behav Res. 2014;49:78–91. 3. Schafer JL. Multiple imputation: a primer. Stat Methods Med Res. 1999;8:3–15. 4. Van Ginkel JR. MI-MUL2.SPS [computer code]. Available at: http://www.socialsciences.leiden. edu/educationandchildstudies/childandfamily studies/organisation/staffcfs/van-ginkel.html. Accessed February 5, 2015. 5. SPSS Inc. SPSS 19.0 for Windows [computer software]. Armonk, NY: IBM Corp; 2010.
Authors’ Response: In our article, we indicated that “the field has not advanced to provide methods for addressing repeated measures ANOVA analysis” and cited van Ginkel and Kroonenberg.1 It should be clarified that this refers to the assertion by van Ginkel and Kroonberg1 that “the implementation of the combination rules needed for ANOVA in statistical software packages has either been quite limited or very well hidden.” It should also be clarified that van Ginkel and Kroonberg1 have proposed methods to conduct the analyses using additional macros and explanations not integrated into major
statistical software packages and argue for the inclusion in future upgrades. As van Ginkel and Kroonenberg1 state: “To date, the implementation of the combination rules needed for analysis of variance (ANOVA) in statistical software packages has either been quite limited or very well hidden. SPSS 19.0 does not provide any pooled results for F tests for multiply imputed data. Using an SPSS macro (Van Ginkel), the procedure may be carried out in SPSS but this still requires substantial manual preprocessing. For repeated measures ANOVA, the procedure is even more complicated. Although SAS 9.3 and Stata 10.0 have a procedure for carrying out the combination rules for ANOVA in multiple imputation, the manual provides no clear practical guidance in how to use them in the specific context of ANOVA. The procedure outlined in this article should also be of assistance to any SAS or Stata user who intends to use multiple imputation in this context. In the appendix, it is explained how exactly the procedures can be carried out in SAS 9.3, Stata 10.0, and SPSS 19.0 using the macro of Van Ginkel. Considering the involved task needed to actually carry out the described procedure, its implementation in future releases of SPSS and other software packages, as well as a clear guidance with examples in the SAS and Stata manuals, seems desirable if its use is to become commonplace.” Richard Solomon, MD* Laurie Van Egeren, PhD† Melisa Quon Huber, PhD† *Ann Arbor Center for Developmental and Behavioral Pediatrics Ann Arbor, Michigan †Michigan State University E. Lansing, Michigan REFERENCE 1. van Ginkel JR, Kroonenberg PM. Analysis of variance of multiply imputed data. Multivariate Behav Res. 2014;49:78–91.
www.jdbp.org | 225
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
