Analyzing behavioural data

Question

Background and experiment details

We have just completed data collection for an experiment. The experiment was a long term study which involved assessing the effect of a drug on behavioural attributes.

Behavioural data involved collection of over 22 attributes/characteristics measured every 10 minutes, for 50 minutes a day for 5 days per sample (so in short we have 25 readouts per sample over the 5 day period).

As this data was collected before the experiment was started, and after 10 weeks of treatment (so basically collect initial data + 1 week + 10 weeks treatment + 1 week + collect final data), so in short We have behavioural data from 2 time points.

Number of samples

The experiment involved 6 groups with variable samples per group. It has been summarized below. Each group had a treated and placebo sub-groups.

----------------------------------------------------------------------------
|      group          |       treatment           |        no of samples   |
----------------------------------------------------------------------------             
         A            |        Treated            |            5           |
         A            |        Placebo            |            3           |
----------------------------------------------------------------------------
         B            |        Treated            |            3           |
         B            |        Placebo            |            5           |
----------------------------------------------------------------------------
         C            |        Treated            |            6           |
         C            |        Placebo            |            5           |
----------------------------------------------------------------------------

Data type

The data is quantitative and a function of time (eg. How long did the subject stay still), distance or counts of activities (eg. How many jumps) etc.

EDIT: Note that the function of time is not time itself, but a number which denotes that. For example a less active subject will have a readout of 2000, whereas a more active subject a readout of 4000 etc. In the end these data points are numbers which can be directly compared to each other. My aim is to see if these numbers are statistically different from each other between the treatment and control groups if tests are performed. Th raw data for in sample looks like this.. (you may have to delimit by space, but you get the idea)

Subject     Total Activity  Fast Activity   Slow Activity   Total Static Counts Fast Static Counts  Slow Static Counts  Total Mobile Counts Fast Mobile Counts  Slow Mobile Counts  Total Rearing Counts    Fast Rearing Counts Slow Rearing Counts Total Center Rearing    Fast Center Rearing Slow Center Rearing Active Time     Static Time     Mobile Time     Rearing Time    Front To Back   Innactive Time  Distance Travelled Meters
2   801 212 589 587 75  512 214 137 77  131 91  40  1   1   0   316 247 69  196 17  284 19.43

I picked a few random samples, averaged the repeated readings and pooled them togther and found out if they were normal. Here are the results

Tests of Normality for random samples ( all 25 readings per sample from 6 samples pooled)                       


Kolmogorov-Smirnova       |     Shapiro-Wilk        
Statistic   df  Sig.  | Statistic   df  Sig.
   .077        123  .068  | .976           123  .026
a Lilliefors Significance Correction

So they are not normal...

Also, The same samples done individually (the 25 readings per sample plottedindividually)

Tests of Normality                          
        Kolmogorov-Smirnova         Shapiro-Wilk        
    ID  Statistic   df  Sig. |  Statistic   df  Sig.
    2.00    .125            20  .200*|  .981            20  .951
    8.00    .255            21  .001 |  .871            21  .010
    17.00   .181            22  .060 |  .900            22  .030
    26.00   .091            20  .200*|  .955            20  .443
    31.00   .148            20  .200*|  .928            20  .140
    35.00   .110            20  .200*|  .926            20  .128
* This is a lower bound of the true significance.                           
a Lilliefors Significance Correction                            

As we can clearly see, some samples are normally distributed and some are not...

Questions we'd like to ask

Null hypothesis is that the treatment produced no change in any of the 22 attributes of activity measured.

We would like to know (assuming the experimental conditions were the same) whether the treatment had any statistically significant effect on any of the 22 behavioural attributes in the following combinations primarily.

Considering only the data collected at final timepoint

• Within the treatment and placebo (in each group)
• Comparing treatment of Group A with placebo of Group B and C
• Comparing placebo of Group A with placebo of Group B and C

Considering data collected at initial and final time points

• comparing degree of change between treatment and placebo (in each group)

So, how do we do it? Our original approach was to average the 25 datapoints collected over 5 days (hence avg the datapoints per sample), and do a t-test between the groups... is that sufficient?

I really do hope I am clear and I am willing to upload more data as and when required