ANOVA and mediation- could someone CHECK over this with me? Help gratefully received

#1
Hi,
I really hope someone could help me out. I'm having a crisis of confidence. I have a PhD in psychology but it's been a while since I did "proper" stats analysis. I would be super grateful if someone could have a look over a completed ANOVA and mediation analysis (simple mediation - dichotomous IV, one M). I'm just not convinced I have reported this correctly and would really appreciate some advice from a generous, lovely person who is more expert than I in these matters!

Please get in touch - feeling a bit stuck so help appreciated.

Thanking you in advance
Faith
 
#2
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

It's a 2x2 full factorial ANOVA and a starter mediation - but it's exciting data with intriguing results so please do get in touch.
 

spunky

Can't make spagetti
#3
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

uhm... maybe you can post it on here for everyone to see so we can all chip in and help?
 

Binka

New Member
#4
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

Ok thanks - So there are a couple of ANOVAs as I have some different "outcome" variables. I'm not sure if I'm interpreting the results properly.

For example - here is the one for a depression outcome -
A 2 by 2 full-factorial ANOVA was conducted with HIV group and gender entered as independent variables for depression as the dependent variable. This analysis revealed that the main effects of HIV group (F(1,393)=20.92, p=0.00001, partial ƞ²=0.05) and gender (F(1,393)=8.083, p=0.00470, partial ƞ²=0.02) were significant. The interaction between HIV group and gender was not significant (F(1,393)=2.210, p=0.13793). Depression scores were lower in those who had HIV and lower in women. The effects (shown by partial ƞ²) are small: only 5% of the variance in depression is attributable to the influence of HIV group and 2% attributable to gender

So because the interaction between group (HIV or control) and gender (male vs female) is not significant, what does that actually mean for the interpretation of the results - I'm finding I'm tying myself up in knots and confusing myself with something that I know is actually fairly simple!

Then I have some other outcome variables/dv - one has a significant interaction (same variables except the DV) is I use p<0.05 as cut off. So for a quality of life variable, main effect group (HIV vs control) sig - F(3,419)=60.489, p=0.000, partial eta square 0.13, main effect gender F(3,419) 26.113, partial eta square 0.059, interaction F=5.501, p=0.019. So how do I interpret that interaction term - does it mean that all four group (male HIV, male control, female HIV, female control) are significantly different from each other?

I'm just so frustrated by myself because I learnt all this stuff so many years ago and now, after years of doing psychometrics (and qualitative, eek) I've just lost the plot with ANOVA! Any help gratefully received !!!
 

Binka

New Member
#5
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

Then there's the mediation analysis. So I have quickly drafted this out and have pasted below. I used the preacher and hayes SPSS macro to do the analysis on pre-cleaned data. I am not sure how to report, what to report etc so I would appreciate any comments on what is below plus any ideas for papers I could look at for well reported (pref psychology type) mediation analyses - I've found a couple but as I'm not sure what I'm looking at it's hard to judge what is "well reported"! As always, your advice and input is very gratefully received. I should say that here we are looking to test whether the impact of HIV group (HIV or control) has an impact on the quality of life over and above the impact of socio-economic status (SES) - so we are hypothesis driven not just fishing! :p

Mediation analysis was completed to explore the degree to which SES (wealth basically) is affecting the relationship between case and quality of life. There are three relevant quality of life variables - which I explain in the text I've drafted in full and dull detail. So I've got a model of IV = HIV group or control group, mediator = SES which is a wealth index and then the DVs are one at a time of the three quality of life variables. (There is a conceptual reason why I want to keep these three variables separate at this time, although there is a potentially more complex story as in reality the three variables do correlate to a small degree with one another).

In all cases, the effect of HIV group (independent variable) on the mediator SES was 1.0869. Table *** provides the summary analysis (I've copied and pasted this below - it looks a bit scrappy though).
In all cases, the relationship between HIV group (case or control) and the WHOQOL domain remained significant when the effect of SES was accounted for. As such, there is no evidence of full mediation of HIV group and QOL by SES.
Although using the Preacher & Hayes (2004) method suggests that partial mediation by SES was significant at p<0.05 in all cases (as zero is no within the 95% confidence interval or CI), the actual effect of SES on the HIV group to QOL relationship is very small. There is no agreed upon way of calculating how much of the variables effect is due to mediation (Hayes, Preacher, & Myers, 2011).

One method is to calculate the proportion of the total effect that is mediated (to create a percentage by dividing indirect effect by total effect and multiplying by 100), as shown in table ***. This illustrates the minimal proportion of the relationship between HIV group and QOL that was found to be mediated by SES.

Table *** Summary of Mediator analysis of socio-economic status and the relationship between HIV case and WHOQOL 1, 2 and 4.
DV Effect on SES on DV Direct effect of HIV group on DV Indirect effect 95% CI of indirect effect Total effect Proportion %
QOL 1 0.0491 1.0404** 0.0100 0.0076 – 0.0007 1.094** 0.91
QOL 2 0.0537* 1.6267** 0.0137 0.0084 – 0.0024 1.685** 0.81
QOL 4 0.0937** 1.6338** 0.0213 0.0103 – 0.0051 1.736** 1.23
* significant at p<0.05 **significant at p<0.01
 

Karabiner

TS Contributor
#6
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

So because the interaction between group (HIV or control) and gender (male vs female) is not significant, what does that actually mean for the interpretation of the results
If there's no interaction then you simply have an additive effect of both factors,
and the effect of HIV status is the same for males and females.
So how do I interpret that interaction term - does it mean that all four group (male HIV, male control, female HIV, female control) are significantly different from each other?
Not necessarily. You could make a graph and inspect the differences between males and
females in the HIV group and compare it with the difference between males
and females in the non-HIV group. Interactions are differences of differences.

With kind regards

K.
 

Binka

New Member
#7
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

Thanks for that w/ ANOVA - anyone ideas re: the mediation stuff above? Much appreciated as I've got as far as I can on my own with this material.
 
#8
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

Any advice on this mediation analysis? I'm just learning this technique so comments gratefully received.
 
#9
Re: ANOVA and mediation- could someone CHECK over this with me? Help gratefully recei

PART 1:wave::yup:

. main effects on depression level in HIV group , p=0.00001 ==> significant difference between depression level of HIV (+ ) & HIV (-)
[II]. gender effect on depression level in HIV group p=0.00470 ==> significant difference between depression level of MALE & FEMALE

above is the interpretation for WITHIN GROUP INTERACTION next we talk about BETWEEN GROUP INTERACTION

[III] interaction between HIV group and gender was not significant p=0.13793) ==> There is no significant difference depression level of male and female i.e. from [II]though there exist a significant difference between depression among male and female still you cannot predict that due to gender there was any significant difference on depression level :shakehead

The interaction between HIV group and gender was not significant


PART 2

You have asked Question about interaction between 4 aspects are significantly different or not
(male HIV, male control, female HIV, female control) = SUBJECTS (say)

here
The Quality of life between HIV (+) and control (HIV (-)) is significantly different
The Quality of life between male & female is significantly different
above are in within subject interaction and result shows that they significantly different

BUT

you have not given the in between subject interaction information Eg P value of interaction between GENDER & HIV (+) / HIV (-)

for this interaction also if P value is less than 0.05 then you may say that all for aspects are significantly different
 
Last edited: