In an experiment, we are testing the effect of a knockout gene on a zebrafish, so we compare knockout vs control (wild type) each group has 8 samples (4 males and 4 females). our goal mainly is to see the difference in the number of eggs produced from each type but also we need to involve and analyze other factors like gender, weight, the age to see if these factors are affecting the number of eggs in the experiment or it's only the effect of the knockout, my data look like the following:
Number of eggs | Gender | Weight | Age
Knockout Male 1
Knockout Male 2
Knockout Male 3
Knockout Male 4
Knockout Female 1
Knockout Female 2
Knockout Female 3
Knockout Female 4
Control Male 1
Control Male 2
Control Male 3
Control Male 4
Control Female 1
Control Female 2
Control Female 3
Control Female 4
So what are the statistics we should do here to study these factors?
You can use paired sample T-test for the analysis like Knockout 1 v control 1 and so on for each of gender weight etc.
Amr Mohamed ,
Good question. There are a few things for you to consider:
1) Experimental Design Effects
If the the assignment of knockout gene to experimental units (each zebrafish?) was conditional on the sex, then, your model should ALWAYS have Sex in the model, in addition, of course, of "knockout" (I will call it "Treatment" now). Therefore, your simplest model would be: Count = intercept + Treatment + Sex
Is this the case for you? Based on how the data is balanced, I would say so.
Regardless of the discussions above, it is required for you to fit these two effects (Treatment and Sex), and there is rarely a reason to split the dataset into smaller ones to perform analyses within groups - we are highly, vastly, extremely advanced ins statistics to do these things....
2) Inclusion of nuisance variables
It is OK, and advisable in many times, to add additional effects in the model that could be accounting for things that could be going to the residual. Thus, if the individual weight or age of the experimental units could have an effect on the response variable (i.e., count), then, sure, why not fitting one or even both?
Generally, age and weight are correlated. Thus, assuming that they impact count, it could be that fitting both you will not find any significant effect on count. However, you may find significance for just one.
3) Type of statistical model
You mentioned that you are measuring COUNTS. Therefore, it is NOT appropriate to use a linear model (such as with t-test). You should use a Poisson model. You can easily fit categorical effects, like Treatment and Sex, and quantitative continuous effects, like age and/or weight, in your model.. that's simple, no problem.
4) A small detail regarding differences in weights between Treatments
If the weights of the experimental units ARE impacted by the treatment, it is not advised to blindly fit it in the model, as the effect of weight would be capturing the effect of Treatment, inappropriately adjusting the count data. There are ways of getting around this, in case yo want to know more about it... But, if Age is not a function of Treatment, then, it might be a better effect to be used in the model.
Please let me know if you have questions. Thanks, Nick
Hello Amr,
Since your outcome variable is count (number of eggs), and, if the "knockout" gene is successful, you'd have instances of zero values, I suspect that a Poisson (no pun intended) regression might be your best bet.
As well, the effective sample size is only 4 per condition, unless male zebrafish also produce eggs. The design appears to be best characterized as independent groups, rather than matched or paired groups, so I would not recommend a dependent/correlated/paired t-test for these reasons.
Are you observing the fish for an indefinite period of time, so that you could obtain data from many spawning cycles, or just for a single cycle (e.g., about a week)? In the former case, the cycles would represent a repeated measures dimension.
My concern about your query is that you list a number of other variables (group, weight, age...) that you'd also like to evaluate as IVs (independent variables). With the sample size you have, you'll have very little statistical power or precision to estimate the degree of influence for all of these. As regards gender, as I said earlier, only the female fish would be expected to produce eggs, right?
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis