It sounds like you have more than two groups. If you have 4 different types of chemicals that you want to compare, and a control, then that is 5 groups. So potentially analysis of variance, with pairwise comparisons of each group compared to the other.

If you then have 4 different days, then you potentially have repeated (serial) measures. You could simply take the average over those 4 days, and that could be valid, but potentially misses something you may be interested in. To take full account of those repeated measures is a more complex analysis that I suspect you do not want to do.

Another issue to consider is that if you are comparing 5 groups, and you are interested in each one compared to each other, then there are potential problems of multiple testing giving you inflated type 1 error, i.e. increased chance of a false positive.

I think that there are many other points that people could make on this, but these are the first ones that come to mind. Good luck.

This is too imprecise: "recorded their behaviour" - what precisely have you recorded?

Apart from this your experiment looks like a reapeated-measures design (guessing from: "recorded their behaviour over 4 different days"). This calls for a repeated-measures analysis. A good suggestion is surely to go for a (linear) mixed model, using the "cell culture" or "well" as a random effect and "treatment" and "time" as fixed effects.

Your research question seems to be (it not very clear from your description!) if the change over time depends on the treatment. This would be reflected by the interaction of teh fixed effects "time" and "treatment" in an appropriate model.

The model needs a defined error structure, and this can be assumed only based on more knowledge about what precisely you measure (counts? proportions? concentrations? ...).

You will likely end up with a generalized linear mixed model. I don't think that Excel provides a (feasable) possibility to calculate this.

I agree with the comments from Darren and Jochen.  You must NOT use a t-test to compare individual subsets of the data otherwise you will most certainly generate Type 1 errors.  The GLIM (generalized linear model) can be accessed through the "R programme' that is freely available on the internet.  A good starting point for this is the book by Andy Hector "The New Statistics with R, An Introduction for Biologists" published 2015 by Oxford Univgersity Press

Dear Darren C Greenwood, Thank you for your reply. I am not interested in comparing each of them with one another. I only intend to compare the different treatments with my positive control. I do not want to compare treatment 2 with treatment 1 or treatment 3 with treatment 2. I studied multiple ways to do the analysis and most of them explain it such that one condition is "Before" and the second condition is"After". As in my case, I have two such situations. 

Case 1: "Before" can be taken as the positive control i.e. before any treatment and "After" can be taken as after the treatments. 

Case 2: "Before" can be taken in terms of time. i.e. Day 0 and after can be taken as Day 1 and similarly day 2 and 3 and so on. 

I am confused about choosing the "Case1" for analysis or "Case 2". I hope I have made my query more clear. I have also attached an excel sheet with the question. If you may, please take a look.

Dear Jochen Wilhelm, I'll be more specific. What i mean is I observed changes in the transcription factors and I noticed that the factors changed in response to different treatments and also changed over due course of time i.e. on different days which you've interpreted correctly but then I am confused about comparing treatments with control or shall I compare Day 0 results with Day 1 results to do a student's t test. Which frame of reference shall I use, Treatment or time? I hope I've put some more light on my problem. Thanks!

 Dear Professor Jarvis, I only intend to show that my findings of different factors in response to different treatments are different from one another and that they do not come from a single "source" i.e. same treatment. This is why I thought that t-test would help me prove this point. 

Dear Udesh, so your read-out is a measure of concentration (of the transcription factor)? In this case you may use a LMM (linear mixed model) on the log concentration.

It is often difficult to find a simple functional relationship between "time" and the concentration (of a transcription factor or alike). If you have such a relationship you are lucky and you can model it (a linear, polynomial, or exponential relationship can be modelled simply using a LMM; a sigmoidal relationship may require to use a non-linear mixed model (NLMM)). If there is no obvious and/or reasonably simple such relationship you can treat "time" as a categorical variable.

You need no "frame of reference" for such models. Bith, "time" and "treatment" are two factors, and their influence (including synergistic or antagonistic effects) on the response (the concentration fo the TF) is modelled.

Forget the t-test. It is not the tool to study such data. Do not think in terms for "how to do a t-test". That's not helpful. Free your mind and ask yourself how to model the response depending on culture, time, and treatment.

http://www.biostathandbook.com/

http://people.virginia.edu/~am3xa/BiostatII/

http://biostat.mc.vanderbilt.edu/wiki/Main/CourseBios312

http://www.bristol.ac.uk/cmm/learning/multilevel-models/

Dear Professor Wilhelm, You are correct that my read-outs are a measure of concentration. If you think that I can treat either time or the concentration as a variable then I think that answers my question. I can compare them accordingly. 

I do not intend to study the data with the software. The expression of transcription factors can be studied through their band intensity after RT-PCR and Agarose gel electrophoresis. I went on and converted that intensity in terms of numerical figures by using ImageJ and now I can study the intensity more accurately in terms of numbers. 

The reason why I wanted to do a t-test is because I think it shows the credibility of a data that it comes from different sources. If I go on and do the t-test and use the results in my publication, it shows to some extent that my results are reliable. That is all why I want to do this test. However, thank you for your time! I'll go through your recommended articles and see if there is anything else tat can enlighten me. Thank you very much!

Do you have access to a statistician at your university? I highly recommend that you talk to one, rather than try and answer this question via this forum. You have been given very good advice, but just as I could not run your experiment, I believe you should get a qualified and experienced person to help you with your analysis.

The mixed model is what I was referring to by "a more complex analysis that I suspect you do not want to do", and these are the methods I teach my postgraduate students. It sounds like you're not comfortable doing that, so you will need to collaborate with a statistician from your institute who knows how to do them. Don't worry - it will give you a nice answer with a nice interpretation.

I also still think that if you want a separate estimate for each of the different treatments with the control, then there is some multiple testing here.

Dear Udesh 

First of all , you need to change your chose of excel , and there are many active programs may be help you better , you must test your data for normality before you decide for parametric test or non parametric test .

I think that T test not the suitable chose because you have 5 groups , and T test used for two independent groups or paired data .

Good Luck 

Hi Udesh  a t test won't give you more credibility if you perform it poorly. No serious journal will accept a paper with a t test to compare multiple groups (unless you present it on different reports). No serious journal should accept a paper analyzed with excel.

As you want to compare treatments only with a control, I recommend you Dunnets post hoc test, beacause it doesn't compare between groups, only between groups and control.

Obviously you should do a previous analysis. The better option could be a mixed model, the second a repeated measures manova, the third an anova with a split-plot design, in wich subplots are the four days. The fourth could be a one way aova day by day, as if were different variables. This last approach can be made on excel, but you haven't access to post hoc tests, unless you program it.

b

What are the questions you want to answer with a statistical significance test.

I can formulate one question quickly by eye balling you r data

H0: the difference between control and treatment X is zero given days

in this case the option would be a paired t-test between control and treatment X where the pair is on day, i e control - treatment X for day 1 then day 2 and so on.

If you want to do an anova as the previous poster mentioned then define your questions first.

All this assumes your data can be approximated by a normal or t distribution,  and I am not familiar with the nature of your data, so follow the practices in your field.

Dear Dr. Niazi, I asked some people around in person about how to tackle this confusion and they suggested the same as your suggestion to compare the control and the treatment individually for each day. Since I only have to two samples to compare, I believe, Excel can do the job. Thank you very much for the help! 

-Udesh.

Dear Udesh,

Dr. Nazi gave correct suggestion when we are testing the significance of several population means  we use ANOVA for it.

In the first place two-way ANOVA is to be done and then if "Fi" is significant then you can go for "t" test

Since you have repeated measures of the same subjects, use repeated measures ANOVA, or, equivalently, a mixed model with subject as a random effect.

Dear Dr. Paul, I haven't contacted any professional statistician yet but I'll use your suggestion and find one. Thanks! 

Dear Professor Greenwoood, I do not work in the field of Bioinformatics and I understand that if I did, things will be much easier for me. However, I think I will find someone who has the knowledge of this field. Thank you very much for your useful suggestion. 

Dear Professor Khalid, I have been suggested by many about not relying on Excel's t test. I think perhaps ANOVA is more reliable for the correct interpretation of the results. Thank you very much!