I am analyzing data from a diary study (multilevel modeling) and based on previous research I want to person-mean center my predictor variables. One of my predictors is work-family conflict (WFC) which was measured on five days. I think I need to calculate the mean of WFC over the five days (for each person) and then subtract the mean from each individual score. This produces new centred scores. But then when I want to create the scale scores each person has a score of zero which I don't understand. If everyone has a score of zero then there is no variance. Am I understanding this correctly?
For anyone looking to person-center predictor variables in SPSS coming here as I did, I thought I'd share the steps as George did for R above...
It's a two step process in SPSS:
1. Create an aggregate variable: the mean score of one person’s observations for the variable. Do this via;
DATA>AGGREGATE in “break variable” move over the participant ID. In “aggregate variable” move over the predictor variable you want to centre (VAR1).
2. Compute the person centred variable from the original variable (VAR1) minus the new aggregate variable (VAR1_mean):
TRANSFORM>COMPUTE VARIABLE: VAR1-VAR1_mean
-
The syntax is:
AGGREGATE
/OUTFILE=* MODE=ADDVARIABLES
/BREAK=PID
/VAR1_mean=MEAN(VAR1).
COMPUTE pmVAR1=VAR1-VAR1_mean.
VARIABLE LABELS pmVAR1 'VAR1 person-centred mean'.
EXECUTE.
-
Hope that can help someone.
Hi,
By scale score I guess you refer to the person average over the five days (e.g. to calculate the routine WFC). This should be calculated from the uncentered daily measurement and not from the person centered variable (if it is calculated from the person centered variable the mean will always be 0 for everyone)
If you want the routine WFC to also be centered you can use grand mean centering.
see for example my paper using this approach with WFC variables here:
https://www.researchgate.net/publication/235009185_The_impact_of_fluctuating_workloads_on_well-being_and_the_mediating_role_of_work-nonwork_interference_in_this_relationship
Article The Impact of Fluctuating Workloads on Well-Being and the Me...
THis may help you to understand
Explaining Fixed Effects: Random Effects modelling of Time-Series Cross-Sectional and Panel Data
Andrew Bell, Kelvyn Jones
Political Science Research and Methods 05/2014; online. DOI:10.1017/psrm.2014.7
and in a person- study see Chapter 15
Developing multilevel models for analysing contextuality, heterogeneity and change using MLwiN, Volume 2 Kelvyn Jones, VS Subramanian
All on Research Gate
Hi George,
Thanks for your response. Yes, the scale scores as you said were calculated from the uncentered daily measurement.
Should I then center those scale scores as opposed to each item from the WFC scale?
For example, if person 1 responds to WFC on 4 days and the scale scores on day 1, 2, 3, and 4 respectively are 3, 4, 7, and 2, should I subtract the average of those 4 days. The average of those 4 days is 4. Therefore, the centered scale scores would be: -1, 0, 3, -2.
If that is correct, is there a way to do this in SPSS by clicking? Or do I hav to use syntax? I can also use R Im just not sure how to calculate these scores.
Thank you so much!
Yes. I think I misunderstood the original problem.
First create the scales (uncentered). Then subtract the person mean to create the person centered variable
If there are missing values e.g. a person responded to 4 instead of all five days then calculate the mean for that person using the 4 scores.
I am not sure how to do it with SPSS but here is one way to do it in R.
For a data.frame called mydata where "pid" is the person ID number and "WFC" is the original uncentered scale:
personmeans
Hello George,
Thank you so much! That is so so helpful. A few more questions if you don't mind:
1)Would you recommend person centering the control variables as well which could technically also be predictors?
2) When I produce the descriptive statistics and the correlation table should I be using the raw scores as opposed to the centred scores?
Thanks for your help!
Also wondering how I can centre all my predictor variables at once without create different data frames each time
1. The decision for person centering any variable should be based on theory - thus it would depend what the control is. However for typical demographic variables you wouldn't need to person center them (e.g. age, tenure, etc) because they wouldn't change over the five days. Thus they would be level 2 controls
2. For the descriptives it is better to use the original variables . (The mean of the person centered scores would be very close to 0). For the correlations it is probably more useful to use the person centered variables (since this is what you are using in your analysis).
In addition, if you are using the person means as level-2 predictors you should provide a different correlation table by calculating the correlations using only the person level data (instead of the longitudinal data). To clarify: if you have 100 participants and data from 5 days, the level-1 correlation table (using person centered variables) will have N=500 and the level-2 correlation table will have N=100.
To person center a number of predictors at once you can alter the previous script.
For example, if you have 10 variables in your dataframe in positions 10 to 19
personmeans
- Badges
- Science topic
- Similar topics
- Psychology
- Organizational Psychology