Rwg(j) with log transformed variables

[Pablo Diego-2]

Hi folks,

Been a passive follower of this listserv for a few years, and appreciating
your collective wisdom, I now come to you with a question:

I´m working on some employee satisfaction data and want to do analysis at
the aggregate level (1 step above - department), in order to look at
relationships with performance indicators at that level. Having these data
(as most satisfaction data) a significant negative skew, I opted for a
logarithmic transformation of my variables before doing any multivariate
analysis. Now I have to aggregate them, and after calculating the Rwg(j)
scores using some cool SPSS syntax I found on this listserv (reproduced
below), I end up with crazy scores for my scales, all of them nearing 1
(see table), even after adjusting for a skewed distribution (the log
transformation didn´t quite finish the job).

         Rwg(j)
Scale 1  0,995
Scale 2  0,986
Scale 3  0,980
Scale 4  0,996
Scale 5  0,988

I must be making some junior Rwg(j) mistake. Are the transformations
messing things up? Shall I just work with untransformed variables when
aggregating data?


Thanks in advance for your help!


compute scale1=mean (item1, item2, item3, item4, item5).
compute scale2=mean (item6, item7, item8, item9, item10, item11, item12).
compute scale3=mean(item13, item14, item15).
compute scale4=mean(item16, item17, item18, item19, item20, item21, item22).
compute overall_scale=mean(item23, item24, item25).
execute.

SAVE OUTFILE='H:\PhD\ORC climate\Merge\Working Datasets\Imputed data\Rwg
file\ind.sav'
 /COMPRESSED.


* aggregating relevant scale items and scale scores in to a group level
* file.

AGGREGATE
  /OUTFILE='H:\PhD\ORC climate\Merge\Working Datasets\Imputed data\Rwg
file\agg.sav'
  /BREAK=yeardiv
  /n=n(yeardiv)
  /xscl1=mean(Scale1)
  /xscl2=mean(Scale2)
  /xscl3=mean(Scale3)
  /xscl4=mean(Scale4)
  /xscl5=mean(overall_scale)
  /xi1 TO xi25 = MEAN(manager1 to overall5)
  /s1 TO s25 = SD(manager1 to overall5).
EXECUTE.

***  scale#1 is items 1, 2, 3, 4, 5.

compute j1=5.
compute a1=5.
compute seu1=1.34.
COMPUTE SXj1=MEAN((S1*S1),(s2*s2),(s3*s3),(s4*s4), (s5*s5)).

***  scale#2 is items 6, 7, 8, 9, 10, 11, 12.

compute j2=12.
compute a2=5.
compute seu2=1.34.
COMPUTE SXj2=MEAN((s6*s6),(s7*s7),(s8*s8),(s9*s9),(s10*s10),(s11*s11),
(s12*s12)).


*** scale#3 is items 13, 14, 15, .

compute j3=3.
compute a3=5.
compute seu3=1.34.
COMPUTE sxj3=MEAN((s13*s13),(S14*S14),(s15*S15)).

*** scale#4 is items 16, 17, 18, 19, 20, 21, 22.

compute j4=7.
compute a4=5.
compute seu4=1.34.
COMPUTE
sxj4=MEAN((S16*S16),(s17*s17),(s18*s18),(s19*s19),(s20*s20),(s21*s21),
(s22*s22)).

*** scale#5 is items 23, 24, 25.

compute j5=3.
compute a5=5.
compute seu5=1.34.
COMPUTE
sxj5=MEAN((S23*S23),(s24*s24),(s25*s25)).
exe.

DO REPEAT rwg=rwg1 rwg2 rwg3 rwg4 rwg5
         /j=j1 j2 j3 j4 j5
         /sxj=sxj1 sxj2 sxj3 sxj4 sxj5
         /seu=seu1 seu2 seu3 seu4 seu5.
compute rwg=(j*(1-(sxj/seu)))/((j*(1-(sxj/seu)))+(sxj/seu)).
END REPEAT.
exe.

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

[Marta Garcia-Granero]

Pablo Diego wrote:
> I´m working on some employee satisfaction data and want to do analysis at
> the aggregate level (1 step above - department), in order to look at
> relationships with performance indicators at that level. Having these data
> (as most satisfaction data) a significant negative skew, I opted for a
> logarithmic transformation of my variables before doing any multivariate
> analysis.
Logarithmic transformation is great for positively skewed data, not
negatively. You are even worsening the lack of normality.


HTH,
Marta García-Granero

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

[Pablo Diego-2]

Sorry, I should have specified I did a reflection before the log, as in:

compute itemlg = Lg10(6-Item ) .

For a 5 point likert scale. I checked skewness after transformation and it was greatly improved, but I am afraid that I won´t be able to use James, Demaree and Wolf (1984) Rwg(j) agreement index with these variables to decide whether aggregation is reasonable...




--- El vie 27-feb-09, Marta García-Granero <[hidden email]> escribió:

> De: Marta García-Granero <[hidden email]>
> Asunto: Re: Rwg(j) with log transformed variables
> A: [hidden email]
> Fecha: viernes, 27 febrero, 2009, 8:31 pm
> Pablo Diego wrote:
> > I´m working on some employee satisfaction data and
> want to do analysis at
> > the aggregate level (1 step above - department), in
> order to look at
> > relationships with performance indicators at that
> level. Having these data
> > (as most satisfaction data) a significant negative
> skew, I opted for a
> > logarithmic transformation of my variables before
> doing any multivariate
> > analysis.
> Logarithmic transformation is great for positively skewed
> data, not
> negatively. You are even worsening the lack of normality.
>
>
> HTH,
> Marta García-Granero
>
> =====================
> 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

      ¡Sé el Bello 51 de People en Español! ¡Es tu oportunidad de Brillar! Sube tus fotos ya. http://www.51bello.com/

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