An insect is put inside a box where it can choose to go to one or another food. The insect can do one choice individually, and several choices are possible. Therefore, the insect cannot do two or more choice simultaneously, but can do one choice after another (different or the same).
The experiment is repeated independently with several insects (replicates).
Thus, the variables in my experiment are:
Individual= random effect (replicate)
1st Choice=Food1, Food2, Food3, Food4
2nd Choice=Food1, Food2, Food3, Food4
3rd Choice=Food1, Food2, Food3, Food4
I'm thinking to use the Binomial Regression (BR) or the Correspondence Analysis (CA). For BR, I would assign 1-value to the Food chosen and 0-value to the others foods. Thus, I would have three dependent variables (0-1), viz. 1stChoice, 2ndChoice and 3rdChoice, and two independent variables, viz. Food (1,2,3,4).
Any suggestion is most welcome.
If insects are moving across foods successively (go after all foods in different order) then I think that a Friedman test is the most suitable.
If this is correct then you should arrange your data like this:
insectA: food1, food3, food2, food4 (hypothetical order)
insectA: food2, food4, food1, food3 (hypothetical order)
etc
insect moved to one food (1st choice), and eventually (but not always) to another food (2nd choice), and eventually (rarely) to another one (3rd choice). 1st, 2nd, and 3rd choices may be different or the same. Insect may do only one choice, i.e. without a 2nd and/or 3rd chooice.
Hi Giovanni,
Just to make it clear, each insect has the opportunity to choose 3 foods, even if it chooses the same food the 3 times, correct?
Also, does the order of preference matters? Or you just want to know at the end which food was most frequently chosen?
I wouldn't go with correspondence analysis because you do have a model to explain the food preference.
I still don't understand which are you independent variables... I can only see the random effect of insect - since each insect will have 3 responses.
I think you can go with an ordinal regression in this case, which 4 categories of response (dependent) variable, as insect as the only (random) effect in the model to account for the repeated measures on the same animal.
Dear Giovanni
I would consider the multinomial choice model that is an example of Dicrete Choice models: http://en.wikipedia.org/wiki/Discrete_choice
... because your answer is not binary but it has 4 alternatives.
If an insect moved to two or more foods in the same trial, I would use as answer the food more consumed or that food where it stay fore more time.
If you consider the order in the preference with 3 replications for individual, you would have 4x4x4=64 differents outcomes. You may study for example if the chance of selecting food 2 after food 1, is the same that that of selection food 2 at first, etc.
I suppose that between evaluation and evaluation. you change the food, and you clean or change the box because of the risk that the previous choices may influence the later choices and to ensure that the results are independent.
May be interesting to weigh how much is eaten of each food?. In this case your outcome would be more informative and you could use additionally other models for continuous variables.
Dear Nick,
insect has the opportunity to choose 3 foods, even if it chooses the same food the 3 times.
I would prefer a model that considers the order of food preference, and the most preferred food.
Independent variables could be three: 1stChoice, 2ndChoice, 3rdChoice, each with four levels (Food1-4). Alternatively, this multinomial variable could be changed to binary.
Actually, these variables are not ordinal, but just categorical.
Dear Guillermo,
multinomial model could be appropriate. I should try it. However, I would use three independent viariables (1stChoice, 2ndChoice, 3rdChoice, each with four levels, viz. Food1-4), to assess to order of food preference. Are you agree? While the use of only one dependent variable (Food1-4), would allow to assess only the most preferred food, but loosing information on the order of food preference. Right?
The experiments were truely independent. In this case the weight of consumed food could not be measured, thus the independent variable is discrete and not continuous.
Dear Giovanni,
Have you considered rank ordinal regression for preference data?
Ana, food choice is not an ordinal variable, but perhaps I cannot understand how to use the rank ordinal regression for my purpose.
Yes, but the preference of your subject will yield data in which subjects ranked your four foods so the ranks will be permutations of 1,2,3,4. You need to put the data in long format so each case shows 4 lines, unless there was missing data for a subject.
This problem is considered from both a statistical analysis and experimental design perspective in the following article:
Taplin, R.H. (2007). Experimental design and analysis to investigate predator preferences for prey. Journal of Experimental Marine Biology & Ecology, 344: 116-122.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics