residualized change scores

[Zdaniuk, Bozena-3]

hello everyone, I would like to calculate residualized change scores and would be very grateful if someone could confirm that I do it right:
1. Regress variable X at T2 on variable X at T1 and ask spss to save standardized residuals (see syntax below).
My understanding is that the residuals would express the change with the linear effect of T1 values removed. Am I correct?
thanks so much!
bozena

DATASET ACTIVATE DataSet1.
REGRESSION
  /MISSING LISTWISE
  /STATISTICS COEFF OUTS R ANOVA
  /CRITERIA=PIN(.05) POUT(.10)
  /NOORIGIN
  /DEPENDENT rpain_intercourse_3
  /METHOD=ENTER rpain_intercourse_1
  /SAVE ZRESID.
=====================
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

[Andy W]

That is correct. Pretty much all instances in which researchers use those residuals in subsequent analyses are inappropriate though, see Gary King's "How Not to Lie with Statistics: Avoiding Common Mistakes in Quantitative Political Science" in which he discusses that. Here is one link that paper is available at, http://www.goethe-university-frankfurt.de/47930041/King_1986.pdf

[Mike]

On Thursday, April 06, 2017 10:56 AM, Andy W wrote:

> That is correct. Pretty much all instances in which researchers use
> those
> residuals in subsequent analyses are inappropriate though, see Gary
> King's
> "How Not to Lie with Statistics: Avoiding Common Mistakes in
> Quantitative
> Political Science" in which he discusses that. Here is one link that
> paper
> is available at,
> http://www.goethe-university-frankfurt.de/47930041/King_1986.pdf

One problem in all this is that a regression on residuals (RoR) may
be appropriate in certain specific situations but, if my reading of
King's
argument is correct, the major difficulty is the "missing variable
problem" that introduces the bias in the analysis.  One way of dealing
with this problem, I think, is to consider the missing variable to
be a mediator and using it (specifically in longitudinal designs)
in the analysis.  But this makes the situation much more complicated
and one needs to decide what assumptions one is willing to make.
One recent article on estimating change effects in a simple
pretest-posttest experimental design (i.e., a 2-way design with
one between-subjects independent variable that represents
treatment and control groups and one within-subjects independent
variable that represents measurement of the dependent variable
prior to treatment and after treatment) with mediation is by Valente
and MacKinnon (2017); see:

Valente, M. J., & MacKinnon, D. P. (2017).

Comparing Models of Change to Estimate the Mediated Effect in the

Pretest-Posttest Control Group Design.

Structural Equation Modeling: A Multidisciplinary Journal, 24(3),
428-450.

They review four different methods of analysis including RoR.
Some might find the analysis that is "best" to be somewhat
surprising.  In any case, reading this article might provide some
food for thought and a further examination of the analysis (analyses)
one may want to do.  Their Figure 2 provides a graphical representation
of the possible relationships among variables.  HTH.

-Mike Palij
New York University
[hidden email]

=====================
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

[Zdaniuk, Bozena-3]

thanks so much to all who responded. I will keep on reading. I was considering using residualized scores because I was asked to use a method described by Judd, C. M., Kenny, D. A., & McClelland, G. H. (2001) (see below for full cit) in which mediation of change in the outcome by change in the mediator can be evaluated by seductively simple OLS regression: The outcome t2-t1 difference is regressed on the sum and the difference of T2-T1 scores on the concomitant  variable. If the sum of T1 and T2 scores is a significant predictor that indicates moderation and if the diff of T2-T1 is significant that indicated mediation .  
The problem I have using this method of testing change being mediated by change is that the raw difference scores are very much negatively affected by the initial scores (higher initial score - less change) and I cannot figure out whether in this OLS method which seems to be using the raw change scores this issue plays a considerable role. Has anyone ever used this method and somehow dealt with the issue of using raw diff scores?
thanks so much for any further directions!
bozena  
________________________________________
From: SPSSX(r) Discussion [[hidden email]] on behalf of Mike Palij [[hidden email]]
Sent: Thursday, April 06, 2017 9:34 AM
To: [hidden email]
Subject: Re: residualized change scores

On Thursday, April 06, 2017 10:56 AM, Andy W wrote:

> That is correct. Pretty much all instances in which researchers use
> those
> residuals in subsequent analyses are inappropriate though, see Gary
> King's
> "How Not to Lie with Statistics: Avoiding Common Mistakes in
> Quantitative
> Political Science" in which he discusses that. Here is one link that
> paper
> is available at,
> http://www.goethe-university-frankfurt.de/47930041/King_1986.pdf

One problem in all this is that a regression on residuals (RoR) may
be appropriate in certain specific situations but, if my reading of
King's
argument is correct, the major difficulty is the "missing variable
problem" that introduces the bias in the analysis.  One way of dealing
with this problem, I think, is to consider the missing variable to
be a mediator and using it (specifically in longitudinal designs)
in the analysis.  But this makes the situation much more complicated
and one needs to decide what assumptions one is willing to make.
One recent article on estimating change effects in a simple
pretest-posttest experimental design (i.e., a 2-way design with
one between-subjects independent variable that represents
treatment and control groups and one within-subjects independent
variable that represents measurement of the dependent variable
prior to treatment and after treatment) with mediation is by Valente
and MacKinnon (2017); see:

Valente, M. J., & MacKinnon, D. P. (2017).

Comparing Models of Change to Estimate the Mediated Effect in the

Pretest-Posttest Control Group Design.

Structural Equation Modeling: A Multidisciplinary Journal, 24(3),
428-450.

They review four different methods of analysis including RoR.
Some might find the analysis that is "best" to be somewhat
surprising.  In any case, reading this article might provide some
food for thought and a further examination of the analysis (analyses)
one may want to do.  Their Figure 2 provides a graphical representation
of the possible relationships among variables.  HTH.

-Mike Palij
New York University
[hidden email]

=====================
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