The FastDay Forum

[carorees]

Ooh, thanks eakman. Please PM me the revised file!

Multiple regression would be the best way to go, so I think we need to recruit more participants than just those who responded to the monthly weigh in? I would imagine that we should exclude those who have been on the diet for less than 1 month though because of the variability in early weight loss?

Thoughts?

Are we not falling at the first fence by picking off two groups - upper and lower quartile - when Kruskal-Wallis (which I am just reading about) seems to like you to have a random selection of subjects from the population ?

Ack, sorry, my bad. Mixed up explanatory and outcome variables. KW is for continuous explanatory variables and more than 2 groups that constitute the outcome variable (e.g. slow, average, and fast losers).

"80% of the biggest loser group didn't count calories on feed days, compared to 30% of the smallest losers. Statistically significant"

That would be crosstables and Pearson's chi square test for independence. Seemingly it also works with more than 2 categories in either variable. However, use increasing categories wisely, as you lose statistical power. You need at least n=5 expected values in at least 80% of the cross table cells. (If not, Fisher's exact may be an alternative)

whereas if we wanted to correlate weight loss against say height we'll have two results as we have two groups. We might then find that weight loss is related to height in the biggest losers but not relevant in the smallest losers"

Correlations, Pearson's in normally distributed, Spearman's if skewed. Scatterplots help to see patterns.

Nevertheless, this amounts to looking at one variable at a time, not combining factors.
Also, if you test a lot of hypotheses, you're bound to "find" some statistically significant associations that may be accidental.

This has already been said (and done): comparing two outcome groups with continuous explanatroy variables needs either T-test (if normally distributed expl. var) or MWU (if skewed).

Generally, in questionnaires and data analysis, when there is perfectly a continuous variable such as age, time on diet etc, it is best to keep it that way, especially during the survey. Mostly because it is very easy to recode a continuous variable into a categorical variable. Similarly, taking arbitrary cut of score or number of participants is not recommended. Instead it is best to benefit from a continuous variable such as weight loss.

Agreed on the possibility to recode continuous into categorical (while you can't do it the other way round). Good point.
However, if correlations are not linear, using continuous variables can be a challenge, requiring transformation procedures etc. Outliers can also mess up the results. I don't know how logistic regression handles this.

Running a series of individual mean comparisons(t-tests, anovas etc...) on different variables between any groups increases the error rate (i.e. p values)

Agreed. I posted my previous reply (where I also say this)without reading this, sorry. However, as a start for exploring the data I find these analyses useful.

Multiple regression would be the best way to go, so I think we need to recruit more participants than just those who responded to the monthly weigh in? I would imagine that we should exclude those who have been on the diet for less than 1 month though because of the variability in early weight loss?

Thoughts?

You need someone who knows how to do logistic regression (i.e. not me)

You need at least 10 cases per outcome group per predictor/independent variable according to my stats book.

There shouldn't be high intercorrelations between the predictors.

If you have categories in your predictors (explanatory) variables, make sure that there is a reasonable number of cases in each category(i.e. "did you fast while standing on your head? Yes/No" is probably not a good question)

There are too few men - if you want to do even non-regression stats, some serious touting will be necessary.

If you are using weight loss as a continuous variable, you could either control for time on diet (that is the beauty of regression, but it means another predictor variable), or assume that weight loss has stabilized after 4-6 weeks and only use those data.

That caveat about time on 5:2 also goes for non-regression analyses.

[carorees]

So, what would you recommend as the best way forward?

Hi Caroline,

Great survey. Can't wait to undertake it.

Q. 12 is a tough one and I don't like to create problems without solutions but I haven't really one for this...Meal delivery programs like Jenny Craig or Diet Factory or Liteneasy and then there are the whole range of other sachet powder meal replacement programs so I don't think you can just name one brand. And then there's the people on the Beer Diet, Cotton Ball Diet, Breatharian Diet - well I suppose that one is low cal...but you know what I'm saying...

There will always be an OTHER category.

I know you want to reduce your response categories.

I just cut and pasted the question here and not sure what c is. Is it a powder?

If the "other" category is too broad, then you are going to get all the way to Little Cats U,V,W,X,Y and Z.

12. Had you been following a weightloss diet in the 6 months prior to starting the fast diet?
a. No
b. Calorie counting
c. Slimming world
d. Weight watchers
e. Low fat diet
f. Low carb/Atkins/paleo
g. Prepared diet meal delivery service eg. Jenny Craig
h. Other

[carorees]

Should it then ask about whether it was high carb/low fat or low carb/high fat or simple calorie counting...i.e. people would need to know the principle behind the diet. I would imagine that the factors that would influence subsequent weight loss when changing from one diet to 5:2 are the degree of calorie restriction and whether it was high/low carb.

The questionnaire has a lot of relevant and interesting questions!

Some thoughts: I would reduce the number of categories for the answers when coding them. Some answers like body fat will be unreliable and also create a lot of missing data, making your results less reliable. Instead of apple/pear I would prefer waist/hip ratio, seems somewhat more reliable to me (I may be wrong). Instruct participants to calculate their TDEE on a calculator of your choice. There are bound to be intercorrelations (e.g.post-menopausal female and age, age and weight, BMI and exercise, etc). Calories on fast days may vary. Not everyone may count calories on fast says or be able to answer in this datail on dietary composition. Unfortunately, I think there are far too many questions for a sample this size, regardless of which type of statistics you use. This is a questionnaire for another PhD thesis. I really think you did a great job, but from my research experience (having set up overly ambitious projects and having to reduce them to a few essential items) I don't see how you can look at all of these variables (with a lot of subjective answers) and a sample with 25 to 30 people in each group (not even with 50). I too would like to know the answers to all these questions, and I really hate to say this, but I think both regression analyses and the amount of info your gathering is too ambitious. By all means, others may disagree with my view.

I would stick with a couple of bivariate tests (crosstabs and correlations) in a well-defined group, and regard the results as exploratory and hypothesis generating).
Maybe I'd start off with formulating a few clear hypotheses.

Again, really sorry to be a spoil sport! Other views may vary.

[carorees]

I was not planning on analysing the data from all those questions! My idea is that we should look at the questions that are most likely to give useful answers (i.e., those that are not too vague). I have started a thread on TDEE as I'm not sure which calculator is best, currently I've just got questions in there that can be used to calculate TDEE but you're right it would be easier to get the respondents to do that job for us!

I'll trim the questionnaire down, in the meantime, hypotheses please...

the BMR equations are simple enough to generate from the weight and other inputs, perhaps we should ask about activity level then generate BMR and TDEE in the results sheet (TDEE being an activity level multiplier of BMR).

Hypothesis - the average feed day calories of the biggest loser group was lower than that of the smallest loser group.

Hypothesis - the average carbohydrate intake as a % of diet was lower in the biggest loser group.

[carorees]

Revised questionnaire here: http://dl.dropbox.com/u/32116230/52fast ... naire.docx

(Thanks eakman!)

I think it is pretty much ready to send now. All we need to do is to finalize who we are sending it to! I wonder if we should just call for people who meet our inclusion criteria (to be discussed a bit more perhaps) to download the questionnaire, fill it in and then upload it. The question is, where should it sit for downloading and where should users upload it? I am a bit of a novice when it comes to cloud storage and sharing! Although I have a dropbox account I only have the free space so don't want it filled up with hundreds of questionnaires! I gather google docs would work but I've never really used it so I need someone to take over here!

I would suggest that once the questionnaires are received and the data input we do some simple stats like chi-squared and see where that takes us.

I can set up a Google mail account for people to email attachments to, probably simplest.

I have a fat dropbox a/c courtesy of my phone but not sure if you can do public upload (potentially risky).

Your spreadsheet format for collecting results worked in google docs ok but I decided most people would be happier editing a word doc and sending it back

[carorees]

I've just added another question about body type...might be a mistake...what do you think?

You can't do a public upload to dropbox but you can allow docs to be emailed I think. If google docs works better, let's use that.

52fastdietforum <at> gmail.com for mailing them in.

Edited to deter spammers, replace the <at> with the proper symbol