Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Does MANCOVA suit my research question?

From   Sascha Struys <>
Subject   Re: st: Does MANCOVA suit my research question?
Date   Sat, 14 Jan 2012 14:10:06 +0100

Dear Cameron, Statalist users,

Thank you very much for your insightful response.
Your Answer provides me with insights and references I would have
surely overlooked otherwise.
I will look into the (latent) residualized change score model and hope
to use it (the outcomes are in fact scales consisting of 11 and 8

There is still something that is not yet clear to me though.
Can MANCOVA be used to examine the association of two continous
variables without comparing groups?
Does the multivariate statistic (e.g. Pillai's Trace) stand for the
amount of variance that an (continuous) IV explains in the outcome
measures when the variance of all other IV in the model is adjusted
for? If not, is there an other way to show if the association of the
continuous predictors with the continous outcomes combined is
statisticlly significant (i.e. some kind of test).

Thank you for your consideration,

S.Y. Struijs

2012/1/14 Cameron McIntosh <>:
> I think you could accomplish the same thing with a residualized change score model, which might be a bit easier/elegant to present and interpret. Just do a multivariate regression of both outcomes on all predictors (including the baseline measure) -- too bad you don't have a longer follow-up period that would allow you to observe and model the longer-term trend in your outcome.If you have multiple observed indicators for your continuous outcome (i.e., if it is some form of summary scale score), then you could also control for measurement error by upgrading to a latent residualized change score model:
> Cribbie, R.A., & Jamieson, J. (2004). Decreases in Posttest Variance and The Measurement of Change. Methods of Psychological Research, 9(1), 37-55.
> Ferrando, P.J., & Anguiano-Carrasco, C. (2011). A Structural Equation Model at the Individual and Group Level for Assessing Faking-Related Change. Structural Equation Modeling, 18(1), 91-109.
> Leonhart, R., Wirtz, M., & Bengel, J. (2008). Measuring effect sizes using manifest versus latent variables: consequences and implications for research. International Journal of Rehabilitation Research, 31(3), 207-216.
> Vautier, S., Steyer, R., & Boomsma, A. (2008). The true-change model with individual method effects: Reliability issues. British Journal of Mathematical and Statistical Psychology, 61(2), 379–399.
> Steyer, R., Eid, M., & Schwenkmezger, P. (1997). Modeling true intraindividual change: True change as a latent variable. Methods of Psychological Research–Online, 2, 21–33.
> Mun, E.Y., von Eye, A., & White, H.R. (2009). An SEM Approach for the Evaluation of Intervention Effects Using Pre-Post-Post Designs. Structural Equation Modeling, 16(2), 315-337.
> Cam
> ----------------------------------------
>> Date: Fri, 13 Jan 2012 20:08:23 +0100
>> Subject: st: Does MANCOVA suit my research question?
>> From:
>> To:
>> Dear Statalist users,
>> I have doubts regarding the suitability of the statistical model I am
>> currently using to answer our research question. I am using Stata
>> 11.2.
>> The following variables are involved:
>> - Two continuous outcome measures measured on two occasions (baseline
>> and 1-year follow-up) Out1, Out2, Cov1, Cov2
>> - Four continuous predictor variables, Pre1, Pre2, Pre3, Pre4
>> - Four continuous, one ordinal and three nominal covariates, Cov3,
>> Cov4, Cov5, Cov6, Cov7(ordinal), Cov8(nominal), Cov9(nominal),
>> Cov10(nominal)
>> (these are demographics and measures that are associated with the
>> outcome measure)
>> We want to estimate the association of the four predictor variables
>> (Pre1 to 4) with the two outcome variables (Out1 & 2) assessed at the
>> 1 year follow-up,
>> adjusting for the same measure assessed at baseline (Cov1 & 2) and
>> other covariates (Cov3 to 10).
>> Next to that we predict that the four predictor variables will be
>> stronger associated with outcome 1 versus outcome 2.
>> I took the following steps: first, I build a MANOVA model testing for
>> the overall significance of the model
>> and the overall significance of each predictor variable on both
>> outcome measures combined.
>> Second, I performed multivariate regression analysis to obtain the
>> coefficients of the predictor variables for both outcomes.
>> Third, I performed four F-tests to test the hypothesis that the
>> predictor variables are stronger associated with outcome 1 versus
>> outcome 2.
>> I used the following syntax:
>> . manova Out1 Out2 = c.Cov1 c.Cov2 c.Cov3 c.Cov4 c.Cov5 c.Cov6 Cov7
>> Cov8 Cov9 Cov10 c.Pre1 c.Pre2 c.Pre3 c.Pre4
>> . mvreg
>> . test [Out1]Pre1 = [Out2]Pre1
>> . test [Out1]Pre1 = [Out2]Pre2
>> . test [Out1]Pre1 = [Out2]Pre3
>> . test [Out1]Pre1 = [Out2]Pre4
>> My questions are:
>> Is this a suitable statistical model to answer our research question?
>> Does the multivariate statistic for Pre1 (continuous variable) and its
>> corresponding significant (p<0.05) F-ratio
>> state that the overall association of Pre1 with the combined outcome
>> measures, adjusted for covariates and other predictors, is
>> statistically significant at the p<0.5 level?
>> Does this model automatically adjusts the means of the outcome measures,
>> so that an overall significance of a predictor demonstrate that the
>> effect of that predictor is significant
>> given that the sample would have the same scores on all other
>> covariates and predictors? Would the method of analysis then be
>> Thanks for your consideration,
>> S.Y. Struijs
>> *
>> * For searches and help try:
>> *
>> *
>> *
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index