You need to determine whether your data is paired or unpaired as the statistics is different for the two cases. This might help http://www.tufts.edu/~gdallal/paired.htm

https://en.wikipedia.org/wiki/Paired_data

http://asdlib.org/onlineArticles/ecourseware/Harvey/PDFFIles/PairedExp.pdf

The statistical tests are different depending on whether there is pairing or not.

Eg you have a paired t test

http://www.statstutor.ac.uk/resources/uploaded/paired-t-test.pdf

and an unpaired t test

http://www.statstutor.ac.uk/resources/uploaded/unpaired-t-test.pdf

You need to be careful as somtimes you can get partially paired data

https://www.ncbi.nlm.nih.gov/pubmed/25834090

and this can require special methods

https://www-users.york.ac.uk/~mb55/overlap.pdf

or

you may need to use permutation methods

see

http://www.ima.umn.edu/preprints/pp2004/1963.pdf

which is explained with examples here

http://www.resample.com/content/software/excel/userguide/RSXLHelp.pdf

for which you might need software

see

http://www.resample.com/download-excel/

Your data may well be partially paired and partially unpaired if you have taken samples from different animals and if you have taken multiple samples from the same animal.

So first determine the degree of pairing and then decide on what test you need. Do not forget the other things that can be violations of a test such as whether the data needs to be normally distributed, whether the variances are equal or not (all of which depends on the test you use).

You need to determine whether your data is paired or unpaired as the statistics is different for the two cases. This might help http://www.tufts.edu/~gdallal/paired.htm

https://en.wikipedia.org/wiki/Paired_data

http://asdlib.org/onlineArticles/ecourseware/Harvey/PDFFIles/PairedExp.pdf

The statistical tests are different depending on whether there is pairing or not.

Eg you have a paired t test

http://www.statstutor.ac.uk/resources/uploaded/paired-t-test.pdf

and an unpaired t test

http://www.statstutor.ac.uk/resources/uploaded/unpaired-t-test.pdf

You need to be careful as somtimes you can get partially paired data

https://www.ncbi.nlm.nih.gov/pubmed/25834090

and this can require special methods

https://www-users.york.ac.uk/~mb55/overlap.pdf

or

you may need to use permutation methods

see

http://www.ima.umn.edu/preprints/pp2004/1963.pdf

which is explained with examples here

http://www.resample.com/content/software/excel/userguide/RSXLHelp.pdf

for which you might need software

see

http://www.resample.com/download-excel/

Your data may well be partially paired and partially unpaired if you have taken samples from different animals and if you have taken multiple samples from the same animal.

So first determine the degree of pairing and then decide on what test you need. Do not forget the other things that can be violations of a test such as whether the data needs to be normally distributed, whether the variances are equal or not (all of which depends on the test you use).

Dr Sood is correct. However, the particular test depends on whether you compare means or proportions or......  Of course if you want to compare individuals you wouldn't pool.

Pooling (or composite sampling) is used to achieve sampling economy. Rather than testing every individual primary unit, you will test one or more pooled samples.

A pooled sample will reflect the average (or the weighted average if the mixing is imperfect) of the variable. For example, factors such as the test specificity and sensitivity are generally different when using grab or when testing pooled samples. Therefore, it really depends on what kind of statistical analysis are you doing.

Helen,

As others have said, it would help to know more about your analysis. For most analyses, though, you would also want to take into account the nested/complex sampling of samples within animals, to account for the part of the variance in the outcome that's attributable to being from a specific animal. In a regression context these are often called multilevel, mixed regression, hierarchical, or random effects models (though be aware that "random effects" and "hierarchical" both have multiple meanings).

Pat

Helen,

One wonders why you are pooling! In general pooling reduces sample size which might seem a disadvantage. But pooling also has the advantage of doing away with individual effects when you are more interested in the conditions these individual animals that are pooled into one measure have in common. In that case the pooling is mainly meant to increase the quality of measurement by removing part of the variance that is considered irrelevant. This means that the actual sample is about the grouping conditions and not about the individual animals pooled together. So, the basic question to be considered is: what are your sample procedures.

The ideal would be to conduct two-stage sampling: a sample of locations where you select your animals followed by selecting some animals within each location. After taking the average (or whatever pooling method you use, it could be a robust measure like a median) these averages can be treated as individual cases in any normal statistical procedure and do not require special formula's or estimation methods from there on.

If however, you decide also to investigate the effects of the the pooling, or the individual variation between individual animals, you have to consider  the suggestions mentioned by Patrick Malone .

By the way, in the past, another reason for pooling was to simplify statistical calculations: by combining the sample into groups the calculations could be done more easily with hand calculators. To correct for this simplification older textbooks often show special formula's that correct for the effects of such groupings on the variances. But this is not relevant in your case.

Helen

How many groups you have, because T test used for two groups only and for normal distribution data.

About data collect for individual animal or group of animals depend on the trait and your planning of experiment , you can use individual animal as experimental unit ( for trait like body weight but you can not use for percent of mortality ) and you can use group of animal as one experimental unit ( pen of 30 hens use experimental unit to study percent of Hen day Egg production ).

Statistical analysis can be performed in both cases , as one observation for experimental unit or many observations for one experimental unit .

Good Luck

Yes.  If  individual animal samples are kept separate you can estimate inter animal variation which is often larger that your treatment variance and it has many implications and uses.