Mixed-Design ('Split-Plot') ANOVA - SPSS (Part 2)
HTML-код
- Опубликовано: 22 авг 2024
- I demonstrate how to perform a mixed-design (a.k.a., split-plot ANOVA within SPSS. I emphasize the interpretation of the interaction effect and explain why it does not matter if the groups differ at time 1 in this type of design.
SPANOVA
This Video is close to 10 years old, but it is still soo helful.
Thank You!
I just wanted to thank you for putting up these great videos. Your walkthroughs were very helpful, and saved my butt...!
Your two videos on this are excellent . I was struggling on internet since morning to 4am next day on how to interpret interaction effects. because my intervention program produced significant main effects but insignificant interaction effects(there was no control group but two ex groups). now got an idea to certain extent. thank you very much..
If you have a 2*2 design, then you know where the interaction effect is taking place, as it can only be taking place from time 1 to time 2 (and between the two groups). However, if you have more than two levels in your design, then you'll have to decompose the interaction effect by doing a series of 2*2 split-plot ANOVAs to see which ones are significant.
Great explanation and interpretation! Very understandable and helpful, thanks!
Oh, and as for the Split-Plot profile plot: Graphs/Line-multiple lines. Set color by group factor, take all the repeated measures (times etc) into Y axis and in Element Properties click on "Display Error Bars". After creating the plot, you can edit it and add markers ("Add markers" option near the interpolation lines).
Great video! Helped me with my MSc research report, Perfect!
Thank you for rescuing my uni assignment!
Regarding whether the effect is due the intervention or the control group. Statistically, this means the same thing. Whether the intervention increases/decreases or the control group increases/decreases is not a statistical consideration; it's an interpretation consideration. Sometimes the control group decreases but the intervention group doesn't change from time 1 to time 2. That's still an effect. Whether it is supportive of your hypothesis is a qualitative consideration.
Thanks for the video. I think you have not explained how to follow up significant interaction. How do we know that one group differed but the other one didn't between time 1 and 2? I know we can look at a graph, but where is the significance value for each line? Thanks!
Hi, first of all, i believe it is just sheer beautiful in what manner you explain statistics!!! Thank you!!! You state that it is in such cases meaningless to interpret the main effects if the interaction is significant. What if the main effects are insignificant, do you believe that one can then interpret the interaction (given the interaction is significant). As in a cross-over interaction, for example. I would be very curious to learn your opinion on that. Thank you very much.
Thank you very much for your series of instructional videos. I find your explanations both concise and theoretically interesting.
Regarding split-plot analysis: We can see T2 interaction is sig diff. However, does this data output actually explain the amount of influence of the intervention, i.e., its moderating effect?
In other words, although we are looking at means of two groups at T1, is it possible to extrapolate at what baseline level moderation appears strongest?
These issues would seem to be even more important when controlling for covariates.
(BTW, is it possible to perform a split-plot ANCOVA using this method, by adding covariates and doing homogeneity of regression tests?)
Apologies for this seminar-style question.
Many thanks!
thanks great explanation. Im glad I found this video.
Thank you so much for your great video. I'm wondering if it is possible to ignore significant Box's M test result when cell sizes are equal. Also, I have significant result of Levene's test of equality of error variances. What should I do with this issue?
Very helpful, thank you !
Great video! Can I ask a question though.
When you have a significant interaction, this tells you that there was a significant difference (over-time/between pre- and post measures) between (only) one of the groups, but not both of them?
If this is correct, and when the interaction is significant, how do you know for which groups the significance is applicable? (For which group was the change over time significant).
I hope that makes sense - thanks again for the great video.
Fantastic Video! I am wondering whether you have any advice on what do say if there is a significant interaction effect but not a significant main effect for group? The video is great, thanks again.
My 3-way interaction in the Within Subjects Effects table is not significant, but the linear effect in the Within Subjects Contrast table is significant. What's the difference? Can I still use the profile plot to interpret the linear effect?
Your videos are incredibly helpful -- thank you!!
THANK YOU
Thanks for the video!
Could you please elaborate me with the difference between 'Within Subject Effects' of resilience * group and 'Between Subject Effects' of group?
I have three repeated measures and two groups (Control and Treatment).
To evaluate the significance of the treatment, should I look at the reapeated_measure * group or the group significance?
My professor taught us that when we have an interaction, we should split the file by groups and run paired sample t-tests for each group. What do you think about this?
THanks I am starting to see the light at the end of the tunnel, BUT I still need your expertise, I will try to be brief. My experiment has 14 participants (7m and 7f), they were exposed to 4 different lights conditions (2 times each condition), lets say A,B,C,D. Our Ho is that their performance (in a numerical test) will remain the same in the 4 conditions, the Hi is that they will improve performance in B and or D conditions compared to A and C respectivelly. It was also measured heart rate, EEG and Pupil Size. It seems this is the ideal test for my analysis or am I really lost in the tunnel yet? Thanks for your time, I hope you can answer.
Hello, thank you for the video.
If I have more than two groups and more than three time points, how would SPSS do the post-hoc tests?
I mean, would it be within-subject post-hocs or between-subjects post-hocs? Is there some tutorial you could recommend me on the matter?
YOU ARE A GOD. But can you cover how to split the interaction and when do you split it and when you run separate one-way ANOVA analysis? Thank you very much :)
I'm not aware of any techniques that can accommodate such a violation. Some people might do an ANCOVA instead, but that analysis assumes homogeneity of regression slopes. Same with a gain score analysis. It's possible the correlations between the groups are different because the distributions are different. So perhaps you could try a log transformation of the data and then re-do the mixed-design ANVOA on the transformed data. You might satisfy Box's M in this case.
Thanks for these videos. Could you please tell me how you run the post-hoc analyses on the split plot anova? Aside from the visual representation there is nowhere in this example where it shows that it is the treatment group that has the increase. In my example, I have four groups (one between groups factor) and two levels (pre- post) Thanks
well interpreted!
@how2stats Thank you for the explanation.
Thank you a lot!
also, do we look at the "Sphericity Assumed" results on the "Tests of Within Subjects Effects. thanks
Thank you so so so much!!!!!
I did the same analysis with two timepoints and 4 groups for my master thesis. Is there a way to get the simple slopes on these seperate lines for further analysis?
My analysis shows no significant effect for the variable time alone, but when I do a T test to compare the means between the 2 times I get a significant result. Why is that?
What does the box "time" (here called pre_post resilience) mean? He says in the video that both groups combined the resilience increases..can someone clarify that for me please?
How do you go about this sort of test if you violate assumptions... is there are non-parametric equivalent?
thaks! very clear!
I'd say so, if the ratio of largest to smallest variance is less than 4. You can cite Howell for that.
I would like to ask, what should be done if the Box's test is highly significant (p
hi i have 4 groups. 2 of them are 5 grade, 2 of them are 6 grade. i implemented a pretest and a posttest to all of them. i ve chosen one of the 5 grade and one of the 6 grade to be control groups and others as experimental groups. i am trying to find out how did groups affected by exprement and grade. after entering the values what should i do
This is good. I've watched a few of your videos now. Just one piece of advice. Please watch your mouth noises, lots of hard P's and T's, lip smacking and licking, breathing and blowing, swallowing and coughing etc. And now that I've told everyone this, they won't be able to think about anything else. (Glass shattering noise (Mosby et al. November 12, 2007)).
So if I have a 2x2 mixed design (each IV has 2 levels), and I've got a significant main effect and an interaction effect, I do not need to follow up with post hoc tests? It even looks like SPSS prevents me from doing it since my "IVs have less than 3 levels".
What would it mean if your Box's test of equality of covariance matrices IS significant? Can you then still use the output of this test? Or how do you proceed to interpret?
In basic terms, it means that the standard error(s) of the difference between the means is not equal across your groups. The Mixed-Design ANOVA assumes that they are. Box's test is too sensitive, so use p = .005 as your demarcation criterion for significance (rather than p = .05). If it is still significant, you don't have a lot of options, expect possibly multi-level modeling?
Is it possible to use the results from a Split Plot ANOVA if Levenes's test is significant (the Box test is not significant)?
What does the "tests of between-subjects effects" tell you?
Is there a way, by now, to integrate the error-bars within SPSS?
There is in the latest version of SPSS. I'm using the free trial to run some stats and was elated that I had the option of making bar graphs with error bars!...I know this is 2 years too late...sorry!
For my diss. I had to use Excel to make the graphs for my ANOVAs and add in the error bars, so I know that pain.
I want to put your example in citation...i visited your blog but i didnt got any refrence...please provide this
I see your Levene's test is not significant which is good. I'm now running the same test and I got significant Levene's and have no idea what to do..
so Mauchly's test is only important (and only calculated) for more than 2 levels as you said, but you didn't specify 2 levels of what factors, time or group ??
It should be two levels of time - so it's lv1 - pre and lv2- post.
If you have more than 2 measurements under time, e.g., lv1 - pre, lv2-6weeks post and lv3 - 12 weeks post, then Mauchly's test becomes meaningful.
My understanding.
the interaction between the time-variable and a between-subject variable, you cannot do it using SPSS, so how would i do it. I would like to know whether groups 1, 2, or 3 are different at say each time level, lets say we have 5 time levels
I mean group 1 is different to group 2 at time point 1, 2, 3,4 and 5
what happens if our box test of covariance matricies is significant?
It's a common misconception that you need to follow up a statistically significant interaction in a 2x2 design with more analyses. You do not. The ANOVA table interaction effect necessarily implies that the Cohen's d values associated with the two conditions are statistically significantly different from each other.
hi, how do you display error bars on the plot? Thanks!
HOW CAN YOU TELL IF THEIR IS A SIGNIFICANT EFFECT BETWEEN THE 'GROUP' VARIABLE THOUGH?! THANKS!
Look for the 'Tests if Between-Subjects Effects' Table. There's a row labelled 'group'.
how2stats Thank you :)
How do you write up your results in APA format?
are you eating as lots of swallowing going on ?