When using change scores in regression, do I use the difference of Only DV or also the IV?

18 August 2018 3 5K Report

Hi,

I wanted to analyse my longitudinal data (2 time points) with a simple change score in a regression, however, I came up with 3 different models and I am not sure which one to use or why.

I am examining whether EFs are predictive of mathematics. I have several predictors and mathematics is my DV. I collected all the data in time 1 and again the same data set one year later, time 2. These are the 3 options I can see:

1. Take the difference from all variables (DV and IV) and run a regression, examining whether change in my IV, predict change in my DV

2. Take the difference from the DV (mathematics) and run a regression, examining whether the scores from my IV tested at time 1 predict the change in my DV

3. take the difference from the DV (mathematics) and a run a regression, examining whether the scores from my IV tested at time 1 and time 2 taken together, predict the change in my DV.

Could anyone help me in picking the right option and explain why?

Thank you in advance

Michael E Young

Difference scores are the wrong way to go. Read up on Lord’s Paradox.

You can either use an ANCOVA approach in which you covary out the initial score or a repeated measures appproach in which time is used as a within-subjects predictor that interacts with your intervention.

These two approaches do not give the same answer because they answer different questions. If you’re interested in differences in rate of change then the repeated measures approach is the right one.

math = time + EF + time x EF

Coding of your variables will determine the interpretation.

David Eugene Booth

Since you want a predictive model I would recommend the ANCOVA approach rather than repeated measures as described by Prof Young If you really want an explanatory model i would recommend repeated measures see the link and attachment for the full details. Best wishes, David Booth

https://www.stat.ncsu.edu/people/davidian/courses/st790/notes.html

Ujjaval Srivastava

I think that Diference in Difference estimator (DID) will be best choice to measure the change in DV through the change in IV. DID calculates the effect of a treatment (i.e., an explanatory variable or an independent variable) on an outcome (i.e., a response variable or dependent variable) by comparing the average change over time in the outcome variable for the treatment group, compared to the average change over time for the control group.

For better understanding of this method you may refer to:

http://economics.ozier.com/econ626/lec/econ626lecture3_handout.pdf

This link will definitely help you.

What are the best TLR4 & MyD88 antibodies one could use in co-immuniprecipitation (lysed brain tissue)?

Dear all, I was wondering if anyone knows whether MGLtools and Autodock can run on Mac OS?

I got 02 moderator values defining Johnson-Neyman significance regions. Is it an error?

Logistical regression using SPSS for a dichotomous dependent variable, with an independent variable of 1, 2, 5 and 10 seconds?

Do I need to do linear regression before running mediation analysis using PROCESS MACRO by Hayes ?

Do I need to strip the PVDF membrane before staining it with colloidal gold?

PACYC PCR not working after several attempts. Would anyone have any suggestions?

What SPSS analysis should I run with one IV (with two levels) and two DVs?

OpenBCI eeg--has anyone used them?

Is there any expert in Smart PLS who can help me?

How to choose ANN number of hidden layers and nodes?