Analytical Statistics Clarification
April 30, 2004 12:25 PM Subscribe
Psychological statistics experts needed! I have to write my thesis and I need help deciding on the statistics to use for my analysis. I have ideas, I just need clarification {more inside}
(Note: This is a copy of an email I sent to my professors. If it looks funny, that's why.)
I think I need to use a MANOVA (multivariate analysis of variance) in this situation. I have 3 different independent variables which create, I think, a 3 x 2 x 2 factorial design.
One factor is "parent involvement" with low, medium, and high levels. Another factor is "focus of concern" with parent-focus or child-focus levels. The third factor is adolescent ethnicity with either Black or White levels.
All of these factors are between-subjects variables. I do believe that these are repeated measures as each subject watched each video twice, although they were with actors of different ethnicity (once with White actors, once with Black actors) for each of the 6 types of interaction:
- low involvement/parent focus
- medium involvement/parent focus
- high involvement/parent focus
- low involvement/adolescent focus
- medium involvement/adolescent focus
- high involvement/adolescent focus
The factor that we are most interested in is the difference in ratings between Black and White participants. So, we have made that the third factor, creating a 3 x 2 x 2 factorial design.
For each of these factors, subjects rated six questions on a questionnaire. Therefore, we would have twelve items per subject per factor (as we showed each factor twice).
In this case, am I correct in believing that a MANOVA is appropriate? Is there additional information that you would need to assist me with deciding the statistics necessary?
(Note: This is a copy of an email I sent to my professors. If it looks funny, that's why.)
I think I need to use a MANOVA (multivariate analysis of variance) in this situation. I have 3 different independent variables which create, I think, a 3 x 2 x 2 factorial design.
One factor is "parent involvement" with low, medium, and high levels. Another factor is "focus of concern" with parent-focus or child-focus levels. The third factor is adolescent ethnicity with either Black or White levels.
All of these factors are between-subjects variables. I do believe that these are repeated measures as each subject watched each video twice, although they were with actors of different ethnicity (once with White actors, once with Black actors) for each of the 6 types of interaction:
- low involvement/parent focus
- medium involvement/parent focus
- high involvement/parent focus
- low involvement/adolescent focus
- medium involvement/adolescent focus
- high involvement/adolescent focus
The factor that we are most interested in is the difference in ratings between Black and White participants. So, we have made that the third factor, creating a 3 x 2 x 2 factorial design.
For each of these factors, subjects rated six questions on a questionnaire. Therefore, we would have twelve items per subject per factor (as we showed each factor twice).
In this case, am I correct in believing that a MANOVA is appropriate? Is there additional information that you would need to assist me with deciding the statistics necessary?
I agree with jasper411 -- more info is needed here, but I'm guessing you'll end up wanting a repeated-measures ANOVA of some sort. It would help if you could give a list of your independent variables (predictors) and dependent variables (outcomes), as well as a brief explanation of what your hypothesis predicts you will find.
posted by nixxon at 2:14 PM on April 30, 2004
posted by nixxon at 2:14 PM on April 30, 2004
Response by poster: Clarification:
The first two factors (focus and involvement) are manipulated within the experiment. Participants are shown a video where the differences are in level of involvement and focus. Ethnicity is a participant factor, not manipulated by the experimenter.
The 6 variables are different variables. They COULD be combined, but probably shouldn't.
The only reason the measure is being repeated is because this is a follow up study of somebody's prior work. I used their materials and performed a replication study, using an identical procedure as them. They claimed that they did each measure twice to ascertain diversity (each situation is acted out by a White parent-teen pair and a Black parent-teen pair). No treatment or intervention applied.
We showed kids 12 videos. They were like this:
White mom - White daughter - low involvement - parent-focus
White mom - White daughter - low involvement - child-focus
White mom - White daughter - medium involvement - parent-focus
White mom - White daughter - medium involvement - child-focus
and so on.
For Nixxon:
Independent variables:
Focus (parent or adolescent), Involvement (low, medium, high) and Participant Ethnicity (Black or White)
Dependent variables:
6 questions that relate to a child's feelings of closeness to the parent-actor, how much they liked the parent, stuff like that. A 6 question, questionnaire.
Hypothesis:
Based on some past research, Black adolescents will perceive low involvement discipline as "better" as measured by the 6 questions being rated higher than the same questions rated by White adolescents who saw the same tape.
Better? (And, you're right. This is very much too complex. This is for a master's thesis and it shouldn't be this hard. But it is, so I need to deal with it.)
posted by ajpresto at 4:50 AM on May 1, 2004
The first two factors (focus and involvement) are manipulated within the experiment. Participants are shown a video where the differences are in level of involvement and focus. Ethnicity is a participant factor, not manipulated by the experimenter.
The 6 variables are different variables. They COULD be combined, but probably shouldn't.
The only reason the measure is being repeated is because this is a follow up study of somebody's prior work. I used their materials and performed a replication study, using an identical procedure as them. They claimed that they did each measure twice to ascertain diversity (each situation is acted out by a White parent-teen pair and a Black parent-teen pair). No treatment or intervention applied.
We showed kids 12 videos. They were like this:
White mom - White daughter - low involvement - parent-focus
White mom - White daughter - low involvement - child-focus
White mom - White daughter - medium involvement - parent-focus
White mom - White daughter - medium involvement - child-focus
and so on.
For Nixxon:
Independent variables:
Focus (parent or adolescent), Involvement (low, medium, high) and Participant Ethnicity (Black or White)
Dependent variables:
6 questions that relate to a child's feelings of closeness to the parent-actor, how much they liked the parent, stuff like that. A 6 question, questionnaire.
Hypothesis:
Based on some past research, Black adolescents will perceive low involvement discipline as "better" as measured by the 6 questions being rated higher than the same questions rated by White adolescents who saw the same tape.
Better? (And, you're right. This is very much too complex. This is for a master's thesis and it shouldn't be this hard. But it is, so I need to deal with it.)
posted by ajpresto at 4:50 AM on May 1, 2004
Yes, a MANOVA sounds appropriate to me. One thing to be sure about is that that you have enough data points. You typically need a minimum of 20 per cell, so in this case you would need at least 12 * 20 = 240 actual viewings.
I'm not sure how you want to structure the factors, because I'm not entirely sure which adolescents saw which video. As I'm reading it, each student watched two videos, one with white people and one with black. If all other things are equal, (that is someone who watched White mom - White daughter - low involvement - parent-focus also watched Black mom - Black daughter - low involvement - parent-focus), you could nest the subject variable inside the race-video variable and account for a subject effect.
(This basically accounts for individual variability in answers. If one kid answers 6s for all positives and another answers 7s, it will essentially rescale them so that they are comparable)
Considering repeated measures takes into account whether or not seeing the video before affects the scores given. I'm not sure that you could consider it that way because they are not watching the same video more than once, although I have not worked with repeated measures much, so take that with a grain of salt.
It sounds (and again I may be reading it wrong) like you are looking for a significant interaction term between subject race and video involvement discipline.
Feel free to send me an e-mail if you need me to clarify anything I said, as I don't think that was particularly clear.
posted by toothless joe at 8:50 AM on May 1, 2004
I'm not sure how you want to structure the factors, because I'm not entirely sure which adolescents saw which video. As I'm reading it, each student watched two videos, one with white people and one with black. If all other things are equal, (that is someone who watched White mom - White daughter - low involvement - parent-focus also watched Black mom - Black daughter - low involvement - parent-focus), you could nest the subject variable inside the race-video variable and account for a subject effect.
(This basically accounts for individual variability in answers. If one kid answers 6s for all positives and another answers 7s, it will essentially rescale them so that they are comparable)
Considering repeated measures takes into account whether or not seeing the video before affects the scores given. I'm not sure that you could consider it that way because they are not watching the same video more than once, although I have not worked with repeated measures much, so take that with a grain of salt.
It sounds (and again I may be reading it wrong) like you are looking for a significant interaction term between subject race and video involvement discipline.
Feel free to send me an e-mail if you need me to clarify anything I said, as I don't think that was particularly clear.
posted by toothless joe at 8:50 AM on May 1, 2004
Response by poster: Public follow-up: What kind of follow up, post hoc tests should be conducted? Just F tests? Any other thoughts?
Joe, you have email. Thanks.
posted by ajpresto at 9:31 AM on May 1, 2004
Joe, you have email. Thanks.
posted by ajpresto at 9:31 AM on May 1, 2004
This thread is closed to new comments.
The things you are calling factors (involvement and focus), are these attributes of the subjects (sort of like race)? Like are these qualities that are inherent to them (based on some kind of assessment), or do they reflect some kind of randomly assigned experimental condition?
The multivariate part refers to multiple dependent variables - are you treating each question on the questionnaire as a variable in its own right? Or can they be combined to form a scale?
Why is the measure being repeated? Is there some kind of intervention or treatment that you are assessing the effect of?
If you're mainly interested in the differences between Black and White participants, you may be contemplating an overly complex statistical design. What are those other things you're calling factors?
posted by jasper411 at 1:45 PM on April 30, 2004