My friend has obtained spawn data in the toxicological studies carried out for 1 year. The concentration used was different and different chemicals were involved in the experiment. From those environment the data has been achieved i.e., breeding data or you can say total eggs, fertilized and non-fertilized eggs. So can anybody suggest which kind of analysis should be done for such obtained data which can be best for submitting in the scientific work?
Also please let me know which kind of statistical method and software should be used for such fish breeding data.
looking forward for the best suggestions
Thanks
I would assume that your response is # fish born?
You can run a host of Logistic Regressions for your data. You will have 3 columns for each chemical. C1 will be conc of that chemical. C2 will be the number of fish egg that hatched/failed for each conc. The third column will be dichotomous, 1=hatch, 0=fail.
For example your columns will look like:
C1 C2 C3
0 80 1
0 20 0
5 70 1
5 30 0
. . . . . . . . . . .
So, 80 eggs hatched at the control dose. 20 failed to hatch. At a dose of "5", 70 hatch, 30 do not. etc.
Thanks a lot Andrew Ekstrom
please have a look for the attached file and let me know whether you mentioned is the same and will be implemented in the work
For statistical software, I will recommend R-statistic as it is an open source software. However, statistic can't perform any magic if you do not have independent data and sufficient sample size. Hope you have these as your attached file did not suggest that.
Looking at your variables, you need an additional column of chemical type as you used different chemicals (Did I get this wrong?). Hence I will recommend ANCOVA which is a combination of regression analysis and anova.
As a suggestion, variable “number of fertilized eggs” should be considered as a proportion of total eggs.
Folasade Adeboyejo Sir, thanks a lot you are right !! have the independent data and sample size ...thanks for your suggestion.
The answer depends upon two things: 1) what types of data you collected and 2) what you want to show. The lesson that I learned during my PhD was that before you ran the experiment, you need to plan backwards from the data analysis; i.e., you start with the type of statistical test you want to use, which tells you what kind of data you need, which then tells you how you need to set up the experiment.
If time of death was measured, then you want to run a survival function; e.g., Cox regression or Kaplan-Meier test.
If there was only one concentration per chemical, then a t-test should be fine to compare control vs. treatment. One could also use an ANOVA to compare the effects among chemicals.
if multiple concentrations were used per chemical, you can try to calculate an ECx (i.e., the concentration that produces x% response), For that, one can use a logistic regression, but it is generally easier to just use nonlinear modeling to fit a dose-response curve. This can be done using JMP, SPSS, SAS, etc. In R, one can use the drc package by Christian Ritz (see his book "Nonlinear regression with R" for some help). USEPA also has a free and easy to use logistic regression program called TRAP (see http://www.epa.gov/med/Prods_Pubs/trap.htm) and a free but less much less friendly nonlinear dose reponse program called BMDS (see http://www.epa.gov/med/Prods_Pubs/trap.htm).
as suggested by the other commenters, your response in any fertilization type of experiment is best expressed as a %. However, to meet the assumptions of normality in most of the tests above, you will probably need to arcsine(square root) transform the data.
Note that if you want to analyze the effect of the reproduction of an individual fish, you can do that using both the raw number of eggs as well as the % that were affected by the treatment. But, if you use the raw number of eggs, you will have to account for the potential differences in baseline reproductive output; e.g., by running an ANCOVA with body length or condition factor as covariates.
In general, it is best to run all of the analyses listed above to fully explore your data, but you should still plan the experiment backwards from the main analysis that you want to perform.
If your experiment and data are similar to the OECD 210 guideline on early life stage, you may inspire your analysis from what is described there. Typically, as Mark suggested, they propose to perform both logistic regressions and multiple mean comparisons to a control (assuming you have replicates).
- Badges
- Science topic
- Similar topics
- Chemistry
- Biochemistry
- Biochemical Techniques
- Blotting Techniques
- Immunoblotting
- Dyes