I strongly suggest to discuss this with a local statistician.

You will have to clarify what exactly are these "food consumption patterns". What kind of data was recorded and how exactly was it obtained. The study subjects has to be discussed thoroghly to see what the research question is in statistical terms (what is relevant, e.g. is it the late outcome difference, or rather the differences in time-courses, and are there specific models for the time courses?, or is an integral measure of a "cumulatice effect" required, etc etc.).

At the first sight you have a repeated-measures design, so I suppose that your model should account for random factors (schools and/or subjects).

@Emel: There is nothing like "parametric data" or "nonparametric data".

The term "parametric" refers to a statistical model, in which the scientifically interesting effect is represented as a parameter. If effect sizes are ignored, one can address the probabilities of ranks (rank-order statistics). These do not infer specific effects that are represented by model parameters, and therfore such analyses are often called "non-parametric".

I strongly suggest to discuss this with a local statistician.

You will have to clarify what exactly are these "food consumption patterns". What kind of data was recorded and how exactly was it obtained. The study subjects has to be discussed thoroghly to see what the research question is in statistical terms (what is relevant, e.g. is it the late outcome difference, or rather the differences in time-courses, and are there specific models for the time courses?, or is an integral measure of a "cumulatice effect" required, etc etc.).

At the first sight you have a repeated-measures design, so I suppose that your model should account for random factors (schools and/or subjects).

@Emel: There is nothing like "parametric data" or "nonparametric data".

The term "parametric" refers to a statistical model, in which the scientifically interesting effect is represented as a parameter. If effect sizes are ignored, one can address the probabilities of ranks (rank-order statistics). These do not infer specific effects that are represented by model parameters, and therfore such analyses are often called "non-parametric".

Hi Zuraini,

Jochen's answer has important points that will have impact in the analysis. So, before you begin the data analysis, those points should be clear to you.

I'll make some assumptions and suggest a model that takes in account what you describe.

1st assumption) Your outcome (food consumption patterns) is a scalar, interval variable. This might not be the case, if the outcome is a categorical variable.

2nd assumption) Treatment and control conditions were assigned randomly or are ignorable. This is necessary for the control group to provide counterfactual evidence.

3rd assumption) Treatment was assigned at school level, not student level.

4th assumption) Time has linear effects in the outcome (to make the model simpler).

5th assumption) You already have decided which covariates are important.

If those assumptions hold, a multilevel (random/mixed effects) model should be used to analyse the data, as already suggested by Jochen. The estimation of treatment effect should be at school level and the model should take into account individual and school-level variance.

At each time frame, your model might be defined as:

y = alpha + beta * time + e

Where 'y' is your outcome and time is an indicator for each time frame (0 for pre, and 1, 2, 3 for post measures). 'e' is individual time frame error. Remember this model assumes linear time effect, which might not be the case!

'alpha' (the intercept) should have random effects for students nested in each school:

alpha = school + se

where 'se' is student (individual) level error.

'beta' (the slope) should also have random effects for students nested in schools plus an indicator for treatment:

beta = school + treatment + se

(Does research gate support LaTeX typesetting? It would make the equations clearer!)

I know SPSS has routines to fit mixed effect and repeated measures models, but I'm not very familiar with it. I can provide more information in how to fit such a model using R, in case you want it.

But here's a tutorial for SPSS that has a similar design to yours (repeated measures and random effects):

http://www.ats.ucla.edu/stat/spss/library/spssmixed/mixed/mixed_testmkt_intro_02.htm

Dear friends, thank you very much for the input. Yes, mine is a repeated-measures design with all assumptions outlined by Erikson. I will have a look at the proposed analysis and link.

Again, thank you very much.

Hi Zuraini,

Your study sounds interesting - would it be possible to elaborate a little on your outcome measure.  I agree that a multilevel model makes sense in the way described by Erikson, but am a bit confused in terms of how you have measured food consumption 'patterns' as an interval variable?  

Tony

Zuraini,  When you say you are "...examining consumption patterns between two groups"..., you indicate a focus for your inquiry, yet what you focus on in your analysis requires further detailing.  As others have encouraged you to seek local statistical consultation, this is likely due to the need for getting further details about a few things that are needed to select appropriate analytic methods, like: what is the key question(s) you are trying to answer?; what are your hypotheses or predictions for group differences?; then, what are the treatments you have implemented to distinguish the groups, and how was it administered? and, finally, what measures have you collected?   You hint at multiple data collection times, though, depending on your question and hypothesis, which may or may not focus on group differences at each and every point of data collection or which could instead be focused on the aggregate difference across all points of collection, and depending on these choices, different analyses are suggested.  Repeated measures analysis may be useful for the former, but less so for the latter.   So, if you would like more focused assistance, help the talented parties here by detailing some of these questions.  Best of luck.

Kahlid,

sure one could use one-way ANOVA, as one could use it nearly everywhere and anytime, but here this is clearly one of the worst possible solutions. Alike, one could always use a hammer (independent of the problem; you can use it as lever, as paperweight, as projectile, as signal, as weapon, etc etc...).

Thank you for your concern Edwin. Actually this is part of my phd work.

To further clarify my study, what i did was to purposely choose 2 schools with almost similar background. One of the schools was identified as a control school, and the school cooks of another school was exposed to a nutrition module. Based on the cooks' understanding (after receiving the module) and also guidance from the researcher, they were asked to prepare and serve school meals that would suit the children's taste and preference. On the other hand, no interaction/communication related to nutrition etc was done with the control school. The cooks just prepared the meals as per their routine i.e. according to the menu provided by the school.  

It was hypothesized that the meals prepared by the treated school (of whom the cook received the nutrition module) are better and healthier than the control school. The outcome of the treatment were determined based on:

1) food consumption pattern (amount of food consumed by school children of the treated school is higher than the control school)

2) school children's participation (treated school would have higher participation than the control school)

3) BMI

4) Social skills/problem behaviour

5) academic competency

The data were collected before (pre-) and after the intervention (3 post interventions). I have problems with item 1 & 2 since the data were repeated (for 5-consecutive days in each phase) while item 3-5, the data were collected only once per phase. I have analysed the data for item 3 -5 using SPANOVA (mixed/split plot ANOVA) (please correct me if this is not the correct statistical tool to analyse the data).

I hope i have answered some of the questions. Looking forward to having help and guidance you.

Thank you.