Can you better describe this "body movement after the simulated trip".
Hello,
I am a novice in the area of regression analyses, and was hoping someone could shed some light on my situation.
I will be conducting a laboratory-based study on balance control after a simulated trip/slip.
I will have a single measure of the following independent variables for each participant: strength, flexibility, reaction time & age.
However, for the dependent variable of body movement after the simulated trip, I will have 10 trials where for each I will determine body movement after the simulated trip/slip.
Therefore, for each participant I will have a single measure for each independent variable, but 10 values for the dependent variable. I wish to determine if the independent variables are predictive of body movement after a simulated trip/slip.
I guess I am just unsure on how to deal with the repeated-measures aspect to my data. Any input on the appropriate analyses, etc. is greatly appreciated.
Thank you,
-Tyler
Can you better describe this "body movement after the simulated trip".
It seems like you are measuring the same thing ten times [with repetition]. While ANOVA has techniques to deal with this [it has some significant advantages actually] I have never seen these applied to regression. In theory you can do this [because ANOVA and regression are the same method] how you would do it in practice I don't know.
Note that this assumes that you are doing the ten trials essentially at the same time without any intervention between them. If you are assuming some intervention between each trial you probably should do repeated measure ANOVA. If there is significant gaps between the trials I am not sure ANOVA or linear regression would work. You really have a time series then although if a test of autoregression shows no issue you might decide to just do the crosssectional analysis you seem to be doing.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
Some good points noesti.
Are the predictors (flexibility, etc.) going to change between trials?
Thank you for the input thus far.
hlsmith, to answer your first comment, the body movement more specifically will be maximum center of mass (COM) displacement after the participant is perturbed. We quantify this metric via motion capture markers which are placed on the participant's body to get a global estimate of body movement. The final metric is a value in meters.
noetsi, yes I am measuring the same thing ten times. Each participant performs a trial, then there is a brief intermission, then another trial, etc up until ten trials are completed. You're point about how to this method as it is applied to a regression is where I am struggling as well.
Lastly, hlsmith, no the predictors will not change between trials.
Personally I think you would be better off using ANOVA than regression simply because ANOVA seems to deal with these type of issues more often - so you can find comments on issues such as repetition in ANOVA text when you won't in my experience in regression. I also suspect that those who work with this type of issue use ANOVA commonly - clinical trials often work with ANOVA rather than regression if they have a linear DV. But I am anything but an expert in ANOVA.
If you are not assuming this is time series (and you should do a Durbin Watson test) then I would take a serious look at ANOVA. I could not even find anything on line about regression with repetition which again reflects my own experience in this area.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
tylerbweaver (09-22-2015)
What is your hypothesis?
Will the person get better motion score the 10th time because they are familiar with the process, not that anything else changed?
How many people are there?
We hypothesized that a model consisting of said independent variables will be a significant predictor of center of mass movement after a simulated trip.
Participants will have 5 'practice' trials prior to the 10 I have already mentioned, so they should not be getting better during the 10 trials used for analysis. The use of 5 trials as practice is based on prior research in the area.
We hope to collect data from 40 participants (10 per independent variable as a crude estimate), but this is always a concern for this type of research as it can be difficult to find participants who want to willingly be tripped.
Multilevel modeling, which will control for your subject clusters.
tylerbweaver (09-24-2015)
Tweet |