How to normalize my data?

[#1]

Hi there.

So, I have a table with a behavioral rate/ min before and after treatment along 4 different hours for 5 animals. I would like to know how can I normalize this data: I tried to divide each hour after treatment by the mean before treatment of the respective hour. As the answers are strange, it seems that I'm missing something.

What would be the best way to do it?

Thanks,

Alice

[#2]

Take a look at the Distribution Platform.  It has the ability to determine what the best distribution fit is, and then provides options for transforming the data to a normal distribution.

[#3]

When you say '...normalize the data...'. What does that mean? Are you trying to transform the data in some fashion so that it reasonably Gaussian in it's distribution? If so...my next question is 'why'?

[#4]

Good question.  I suspect (though I could well be wrong) that the request is to "standardize" the data rather than "normalize" the data.  If this is correct, then I guess it is because the 5 animals differ enough that the variability across the animals masks the treatment effect.  The usual way to standardize would be to measure how many standard deviations away from the mean each observation is - so rather than dividing the data for each animal by the mean, it would mean subtracting the mean and dividing the difference by the standard deviation for each animal.  Is that the question?

[#5]

My first thought, like @P_Bartell, was, why would you normalize the data? The transformation might lose important information. The variation you eliminate might be informative and explained by changes in another variable.

[#6]

Hi @txnelson, @P_Bartell, @dale_lehman and @Mark_Bailey ! Thank you all for your time and for helping me with my problem.

First, I think I need to better explain what is it about.

I have a result showing the rate/min response of behavior, along the 4-first hours before and after treatment, for five animals. As my results were getting odd, I asked the Professor what would be the best procedure to do this, and he told me to actually consider the rate/min for each animal before the treatment procedure, as my baseline. Without the normalization or standardization procedure, the statists are only showing the treatment effect, but we are wondering if

there is any variability across the animals. Or do you think I'm considering it wrong?

[#7 · Accepted solution]

Add terms to the linear model to address these concerns and test the hypotheses. You can add a term for the fixed effect of each animal. This represents a different average response for each animal. You can add a cross-term treatment*animal for the fixed effect of the interaction. This term represents a different treatment effect for each animal. I am concerned that you might not have enough data to estimate everything adequately.

I think that the animals are simply a sample from a population. In this case, treating animals as a random effect makes sense.

Select Analyze > Fit Model. Enter your response in the Y role. Add treatment and animal as effects. Select the animal effect, click the red triangle next to Attributes, and select Random Effect. The launch dialog looks like this:

[#8]

Thank you, I think that is exactly what I need. And I want be doing any changes in my data.

[#9 · Accepted solution]

To add a bit to @Mark_Bailey 's advice, before modeling I'd spend some time exploring the data graphically. Graph Builder or the Fit Y by X platform are good places to start to visualize the data wrt to the effects you are trying to determine.

[#10]

Thank you for your suggestion, I shall do it first.