QUESTION:
How do I conduct a bootstrapped (within-subjects) one-way repeated measures MANOVA or MANCOVA in SPSS? SPSS has no default setting for bootstrapping these types of analyses. So, in the SPSS data window, is it possible for me to use one column for my Outcome/Independent variable and include the three time points within it (ie. 'Time' as the column title and then its 'groups' as 1, 2 and 3) and then do the same for my 4 dependant variables by putting the individual scores next to their respective time 'group'... So, across the top of my SPSS data window, I would only have 4 variables visible, but within each of those variables would be the 3 'condition' scores. If I then used 'Analyze > General Linear Model > Multivariate' then place 'Time' (with it's three internal levels) into the 'Fixed Factor' box and my 4 dependent variables (with their 3 time/internal measurements/scores inside each single variable) into the 'Dependent Variables' box, the option to use 'Bootstrap' is available in this window. Would this be an appropriate way to conduct my bootstrapped one-way MANOVA? Is this what I would actually be doing?
---Masters level dissertation details are below--- STUDY:
An investigation into the effect of voluntary behaviour regulation upon social media use through smartphones.
AIMS:
1. To establish the effectiveness of 'iOS Screentime' as a tool for behaviour regulation, concerning social media use.
3. To identify whether educational performance 'risk-factors' for young people (derived from the scientific literature) can be significantly reduced through such a 'self-help' technique.
2. To contribute to the debate concerning whether or not social media addiction/reliance is a behavioural phenomenon.
HYPOTHESES (derived from literature):
1. Self-imposed behaviour regulation will significantly decrease the time spent accessing social media, for 'Time 3' and 'Time 2' VS 'Time 1'.
2. Self-imposed behaviour regulation will significantly decrease social media reliance scores, for 'Time 3' and 'Time 2' VS 'Time 1'.
3. Self-imposed behaviour regulation will significantly improve mood scores, for 'Time 3' and 'Time 2' VS 'Time 1'.
4. Self-imposed behaviour regulation will significantly decrease FOMO, for 'Time 3' and 'Time 2' VS 'Time 1'.
IV: Time (Catagorical - 'Time' is my 'grouping variable')
Level1: Pre-Intervention (Time 1)
Follow up - Can I simply create a SINGLE variable for 'Time' coded as 'T1, T2, T3', use that as my 'categorical' IV (which will also reflect my 3 within-subjects conditions) and then run a GLM multivariate analysis against my 3, 4 or 5 variables (depending on what I choose), each of which will have their 3 singular scores across their measurement at 3 different times under 1 variable column in SPSS, rather than their original 3 columns?
Does that work? Robert Trevethan is this something you might be able to help with? (Apologies for being a bother, it's silly AM for me this side of the world, and I'm not sure if I've descended into madness haha)
Alistair Page, I'd love to be able to help you, Alistair, but I'm afraid that I'm a bit bamboozled (well, a LOT bamboozled) by everything that you're trying to do. Certainly, however, I suspect that 13 participants is a very small number to be relying on for valid results to emerge, even if you use bootstrapping.
Do you have access to a tame statistician?
Ha ha, maybe you should write up a master's thesis that deals with the difficulties of obtaining and analysing data concerning certain topics and certain research questions. Sorry, no, that's probably not helpful on my part. However, I'd hate to see you continue to tear your hair out about something that's beyond use of reasonable strategies.
EDIT: I have updated my question above and put the actual question I am having trouble with at the very top. I believe what I am asking makes a lot more sense now. Robert, I would be forever in your debt if you could possibly take a quick look at the amendment and advise?
Robert Trevethan - Thank you for getting back to me, I do really appreciate it! My supervisor is inundated with emails at the moment, so I have found it a little difficult to get hold of him.
I am aware that the data set is not ideal and it is unfortunate that only 13 of my 53 cases made it through to the end. However, in order to attain a solid grade, I need to simply show that my analysis of the data that I have is procedurally correct. Do you think that representing 'time' as a categorical variable (as I have mentioned in my comment on my post) is appropriate? I have read that a sample size as small as 8 can be appropriate for bootstrapping, but I am simply going to cover it as a limitation in my discussion.
All the best,
Alistair
Alistair Page, Hmm, it's sweet of you, but I think you put me on too much of a pedestal with regard to stats. My knowledge about stats hits a ceiling pretty quickly in some areas, and, if I have a weakness, it's not feeling happy about providing advice about things when I know virtually nothing about them. I know, I know, it's a weakness.
As things often go, supervisors are not particularly good at stats, so I do wonder whether you can throw your problems at a statistician - though I'm aware that they are often difficult to come by.
Thanks, however, for letting me know about bootstrapping apparently being permissible with as few as 13 participants. From my heights of naivety, I'm cautious about basing conclusions on only 13 participants, particularly when some distorting self-selection might be behind their participation.
I also wonder whether you have bitten off more than you should be expected to chew at the master's level.
I really regret that I'm not being more helpful at the moment than I've tried to be. Maybe some other, and good, RG members can step in to help.
Robert Trevethan ^_^ honestly, thank you for getting back to me. I have managed to speak with my supervisor and through doing so have actually answered my own question.
I can conduct a bootstrapped one-way repeated measures MANOVA, but ONLY if I can find a plugin for SPSS that will do it, as organising my data as I have proposed in the original question and then using 'Analyze>GLM>Multivariate' and selecting the 'bootstrap' option will actually resample the mean from columns, meaning that, as my three scores across time are within each column, the new resampled means would be worthless (due to the model requiring a repeated measures analysis), so It is not appropriate for a one-way repeated measures MANOVA. If I was only measuring once across each DV, then this analysis would be appropriate.
SO... If I can not find the illusive plugin, then I will be forced to use 4 x bootstrapped one-way repeated measures ANOVA with Bonferroni corrections to adjust my alpha for familywise error. As I will be testing the same data 4 times, I will note this as a limitation of my study concerning the interpretation of any significant findings. Alternatively, I believe a bootstrapped one-way repeated measures MANOVA can be conducted through 'R' statistical analysis software, but I don't have the time to learn how to use it right at this moment.
AND THAT's IT! ^_^, thanks for chiming in and I hope you are well.
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media