I applied an ANCOVA for my analysis and obtained beautiful results but the data were collected in three different batches, that is to say we have collected data from different people three different times in two years, around 75 people each time. I have learnt that multilevel linear models are for hierarchical structures, like different classrooms, schools etc. All our participants are enrolled in the same university. I am not sure if this applies to my situation. What do you recommend, carry on with ANCOVA or apply a multilevel linear model to get rid of the possible effects of different data collection periods?
Thanks in advance!
The time of year may affect the responses. I would add a term for time period and see. There is no reason to assume that participants are clustered within time periods, but measures may be dependent on the time of year you measure them.
Multilevel models are useful if you have a hypothesis in which individual characteristics such as ability interact with context variables such as the school the person attends.
Where you don't have multi-level hypotheses, you can use robust variance estimation to account for clustering of observations within clusters. It has the advantage of working even if the variables that drive the clustering are not fully known.
In my experience people often don't include time period as an extra variable like this, but there's no reason you couldn't do it; in fact, it arguably is better than the usual practice in my field (which is just to ignore it). You can always try making models with and without that parameter and use model comparison to see if it improves the model significantly; if it doesn't, you're safe not using it.
I guess one issue is whether to include batch (or time period or whatever) as a random variable (thus in a multilevel model) or a fixed variable (in which case you could keep using the same model you're already using). Conceptually it seems like a random variable, but I have heard informally that random variables don't really work well unless they have enough levels (like, I dunno, maybe around 15 or more), whereas you only have 3; in this sort of context people have recommended just treating it as a fixed variable instead. Unfortunately I don't have a source to cite for that, it's just something I've heard some stats experts say.
Hello Nehir,
Can you provide us with more informations? What kind of dependent variable do you have? And what are the independent variables? I agree with Ronán and Stephen: mixed models could be useful but we need more information to better help you.
Cheers!
N
Dear Mr. Barbosa,
My DVs are area under the curve with respect to ground cortisol awakening responses (AUCg CAR) for two days. CAR might actually be season and time dependent. My IVs are gender, genotypes for a polymorphism, and a psychological measure (categorized into low and high levels in order to be applicable to ANCOVA). I also have 4 (scale) covariates.
Dear Nehir,
So, you have just one DV and three IV's?
I am not the expert in MLM, but what if you would like to run MLM you need to have multilevel structure. It is recomended to have at least 8 or even 10 objects (in this case: batches) in lvl 2, and at least 10-15 objects (here:participants) in every batche.
75 people is not enough to MLM. It is not enough to have 10 object in lvl 1 and 8 object in lvl 2. The program will not run the anaysis.
Even if you cluster people within time. You have still not enough. 3 measures is still not 8. And what about power effect? Sometimes power effect to show statistical defferences is not enough for even 1000 people in some research.
I would recomend ANOVA :)
In this case I agree with Joanna, maybe the most parcimonious choice will be a simple analysis of variance (ANOVA) for each IV (i.e., gender, genotypes for a polymorphism, and a psychological measure).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics