Please could I get advice to analyse the following dataset. In a trial, animals were provided with feed in different forms (round vs square) and concentrations (100, 80, 60%). This was trialled twice, (Trial 1 and Trial 2- everything was kept similar between both trials). The animals are kept in pens, so each performance parameter (bodyweight etc. is measured as a pen average).
I want to combine the results from Trial 1 and 2 and to determine the effects of:
Please can someone advise if Pen and Trial should be considered as random effects? It is the three effects which I have highlighted that I am most interested in. I have tried running the analysis but my outputs so far do not make much sense so any help would be appreciated- I have JMP, not JMP Pro.
Before providing my opinion to your question, a few things that I noticed on a very quick look at your data.
First, get your data into JMP! Excel files are nice, but JMP files are better, especially when it comes to analysis!
I noticed that you have more than 2 forms of feed. You also have PRILL and POWDER. These look like "add-on" types of trials, meaning they were not part of the original design. I will assume that you will remove these rows from the table as there is not much data on those forms anyhow.
Once prill and powder have been removed, your treatment column is not necessary. It is a combination of Concentration and Form. Therefore, just using Concentration and Form makes the treatment redundant and not needed.
A random effect is an effect where the choice of level is chosen randomly from a larger set. It seems to me that Pen would be random. You have a large selection of pens and the ones used for these trials are randomly chosen.
Trial also sounds like a random effect to me. You could certainly do more trails, so the two that you did run are a "selection" from a larger set.
You state that your analysis did not make sense. The community will be much more equipped to help you if you provide the JMP file with your analyses and with more specific questions on what does not make sense to you.
Echoing Dan, why aren't you attaching the JMP file? In your excel file, there are a number of columns that have not been defined by your brief description. You should define each column (and if it is an x or a y). For each response variable(s) you will need to provide how much of a change in the metric matters from a scientific perspective (practical significance).
After that, we'd be happy to take a look.
I have attached the JMP File- and made the changes suggested.
I have ran a mixed model with Trial and Pen as random effects, then Form, Concentration, Age, Form x Age and Concentration x Age on the data. Every other parameter (uniformity, FCR etc.) is a response variable.
I have ran this on the bodyweight data and the analysis makes sense to me now. Would I be correct in interpreting that for bodyweight:
There were significant effects of everything apart from Age x Form. Out of the two random effects, "pen" had a significant effect, but "Trial" did not.
Is my analysis correct, and is this the correct approach to take, rather than a repeated measures analysis? Is there any further analysis you would recommend?
Abbie, I do not see anything regarding practical significance? How much weight gain is of scientific value? I don't see a column titled bodyweight, so I assume the column weight is bodyweight?
I'm really not sure what questions you are trying to answer or what hypotheses you are investigating. I'm not really sure how you are measuring weight? I seem to recall you are averaging weight per pen...If you are going to calculate an average, you should also consider the variation within pen. Not sure average is the appropriate statistic? How many animals are there per pen? What if you have one really large animal and the rest small, then the average is not an appropriate measure of central tendency.
Some of my thoughts:
1. It seems reasonable that animals will gain weight with age. What is the predicted amount of weight gain? Or perhaps, what is the historical amount of weight gain with age?
2. So is your question: Does the concentration or form of the feed affect weight gain (over and above "normal" weight gain)?
Plotting the data is a great way to start. I have attached a simple chart showing the weight by trial and age, coded by pen.
Weight vs. Age (days) separated by Trial
Obviously you have pen-to-pen variation (there may be a systematic pattern here).
Looking at that variation (within treatment so not influenced by concentration or form)
R of weight within treatment
So obvious age differences and pen-to-pen variation. Does not appear to be much due to the factors you manipulated, concentration and form. This could be due to the level setting of those factors (not sure why you chose 3 levels for concentration unless you predicted some non-linear effect) or they don't have an effect.
You can do fit model by age to remove this huge effect.
Thanks for the response.
There were 318 animals per pen, each week 32 birds per pen are weighed. This gives the bodyweight "weight" column.
The hypothesis I want to test are:
Does feed form (round versus square) affect performance (weight, FCR, uniformity, weight gain and feed intake).
Does feed concentration affect performance (as above).
There are 3 levels of concentration because there were three different levels applied (100, 80 and 60%). In total there were 6 treatments:
1. 100% concentration round pellets
2. 100% concentration square pellets
3. 80% concentration round pellets
4. 80% concentration square pellets
5. 60% concentration round pellets
6. 60% concentration square pellets
The same trial was repeated twice, with this setup- and everything was kept identical. What I want to measure is the effects of feed form and concentration on bird performance from both trials combined, whilst accounting for variation between trials. I would like to know do the factors affect performance, and if so, then what ages this occurs.
Funny how when these analyses start, there is always information not reported...Please keep in mind, that what analysis tools, what questions you can answer, what conclusions you can draw, your ability to extrapolate results, etc. ALL DEPEND ON HOW YOU GOT THE DATA!
The 32 sample of birds is a random sample of the birds within each pen. You are not actually tracking the same set of birds over time. Is your measurement system precise and stable? Did you ever measure the same bird at the same age more than once?
First I would determine if the variation you see in the response variables was of practical significance. Then I would do multivariate analysis of the multiple Y's. Do you expect any of the Y's to correlate (or not)? Why?
Weight is highly correlated with 1. Feed Intake, 2. FCR 3. Weight gain and 4. Uniformity though some of the patterns in the data are of interest.
Color/Mark by Column: Age exposes the relationships are significantly impacted by Age.
Doing the multivariate analysis BY AGE, may be more insightful. Example for Age = 7
Notice above there is little relationship between Weight and Feed intake. And Weight and FCR are now negatively correlated.
I understand you tested concentration at 3 levels, but not why you did this? You have more information about concentration than you do about form. Not a bad thing, just need to realize the study is biased.
You do have some trial-to-trial differences (particularly regarding variation within treatment). You say the trials "everything was kept identical". This is, of course, is impossible. The key is understanding what changed between trials. I think you should treat trial as a block in the analysis. And since we don't know what changed between trials, assign it a random effect in the model. Since you likely have some knowledge that birds will increase in weight as they age, the questions are how do the factors (concentration and form) impact performance (Y's). If you had some historical data: weight over age, you might be able to look at the delta between "historical" and treatment effects. You might be able to look at the rate of weight gain (slope of the line). Also it looks like effects are different as the birds get older.
I would try fit model BY AGE.
There are no labels assigned to this post.