topic Choix analyse pour tester effet d'un traitement in Discussions

Choix analyse pour tester effet d'un traitement

Nicoletti29 — Fri, 09 Jun 2023 00:40:16 GMT

Bonjour,

J'ai des données mesurées sur plusieurs animaux, réalisées à différents niveau de l'aorte (en pj). Un groupe a été traité (DT), l'autre a reçu un placebo (PBS). Je voudrais mesurer l'effet du traitement en prenant en compte les différents niveaux de mesure (distance from start (mm)). J'ai des données manquantes (pas le même nombre de niveaux analysés). Quelle analyse dois-je lancer dans JMP?

Merci d'avance, je suis bloqué...

Re: Choix analyse pour tester effet d'un traitement

ron_horne — Mon, 18 Oct 2021 13:43:18 GMT

Hi @Nicoletti29 ,

perhaps it is worth you try this approach:

it will run a mixed model where the mice are random effects. I am just not sure about the full factorial approach that would use the distance as random as well.

it should look something like this:

and

let us know if it helps.

Re: Choix analyse pour tester effet d'un traitement

Nicoletti29 — Thu, 21 Oct 2021 10:22:33 GMT

First of all, thank you so much for taking time to reply to my request. I have tested the proposed solution and I think that the full factorial repeated mesure is a correct way to analyze these data. Given the small size of my groups, I was wondering whether you would advice to go for a nonparametric test, and which one would be adapted for the structure of my data?

Regards

Re: Choix analyse pour tester effet d'un traitement

ron_horne — Fri, 22 Oct 2021 08:33:34 GMT

Hi @Nicoletti29 ,

i am not sure about how to control for the distance (within subject effect) when doing a non parametric.

the non parametric tests are all available in the fit y by x platform. if there were very few categories of distance you could use it as a by variable or interact it with the treatment to produce one factor with all the combination of categories. In this case, i think the distance is more a continuous variable with many categories.

furthermore, i wonder if the missing observations impact the measurement itself. if there is any dependency between the number of measurements and the results than this makes things even more complicated and i would think you need to introduce another within factor - measurement number. in the past i was dealing with animal data where each measurement by itself would impact the following one. this would happen since the animal was getting used to the person performing the measurement and it would improve over time just by taking more measurements. Therefore, the missing observations also had less opportunity to improve their measurement by acclimatizing to the operator.