I am struggling to decide which statistical test to use. I'll give a short explanation of my performed research:
I've monitored the leg temperature of a chicken for 6 days. On the 6th day, I introduce a stressor (a loud sound). Previous research showed a significant temperature drop at the leg using infrared thermal imaging, when a stressor is introduced.
Since my temperature data is collected using a wearable leg sensor, temperature data is highly contaminated. Therefore, I automatically detected 5 distinct periods (e.g: sitting periods are characterized by a remarkable increase in temperature since the chicken is sitting on the sensor, and hence increases the sensing temperature). For these 5 periods (sit, stand and heating, stand and cooling, activity and heating, activity and cooling), at least 15 observations are available. So I have the means and standard deviations for each of these 5 periods.
I want to detect if the continuous temperature data of a stressor-period significantly differs from the continuous temperature data of the 5 other periods.
Unfortunately, I only have 1 measurement of the stressor-period... I was thinking to perform a (multiple) paired t-test between the 5 non-stressor periods and the stressor-period, but since I only have 1 measurement of the stressor-period, I don't think that is feasible... Can anyone help me out or suggest some literature I should look into?
I'm glad that you turned to this forum for help. We will try!
Your study is to confirm that the stressor causes a decrease in the temperature. I assume that you have multiple subjects. Then a one-sided paired t-test is appropriate.
You believe that the data are "highly contaminated." I am not sure what you mean or if this claim invalidates the data. You continue to describe how the temperature reading depends on the position of the subject. Is that what you mean by contamination?
You measure the temperature during 5 distinct periods prior to the treatment. Does one of the periods match the position when the stressor is applied? If so, then this period is the only one relevant in the matched pairs test.
Why are you able to make multiple measurements before the treatment but only one measurement after the treatment? Also, what do the multiple measurements represent, replication or time series or something else?
Am I even close to understanding the issue?
Thanks you for your fast reply!
"You believe that the data are "highly contaminated." I am not sure what you mean or if this claim invalidates the data. You continue to describe how the temperature reading depends on the position of the subject. Is that what you mean by contamination?":
Yes, at specific moments in time, temperature rises or drops during specific behaviors. During sit, temperature will always rise. While in standing position, temperature sometimes rises or sometimes drops. The same is true for activity behavior.
"You measure the temperature during 5 distinct periods prior to the treatment. Does one of the periods match the position when the stressor is applied? If so, then this period is the only one relevant in the matched pairs test.":
Thank you! This is the one I was looking for I guess... When one of the 5 periods matches the behavior when the stressor is applied, I could perform a statistical test with the null hypothesis being 'the temperature during stress-period does not differ from the temperature of the matched behavior'. Right?
Why are you able to make multiple measurements before the treatment but only one measurement after the treatment? Also, what do the multiple measurements represent, replication or time series or something else?:
So, i have 12 chickens, 4 groups, 3 subjects per group (each subject has a sensor). A testweek consists of 7 days. 3 days familiarization to the sensor device, and 4 consecutive testing days where 4 different stressors are applied (1 stressor per day). During these 7 days, I extracted several instances of the 5 specific behaviors, so I have more than 15 observations each for the 5 specific behaviors. Lets say, on the 4th day, a specific stressor (e.g short, loud sound of 2seconds) is introduced. I only have 1 measurement of what happens with the temperature per chicken. Since (baseline) leg temperatures differ a lot between chickens, I can't compare the 'stressor-period' of chicken1 with the 'stressor-period' of chicken 2 or 3... So, I only have one measurement post-stressor for each chicken, while I have multiple measurements of the 5 specific behaviors for each chicken.
Thanks for helping me out!
Yes, use the activity that matches the activity under stress for the null in a matched pairs (before, after) analysis.
Okay, thank you very much!
But what with the issue of only having 1 measurement for the stress-period? This stress-period is continuous data, should I just compare the mean and the standard deviation of this single stress-period with mean and standard deviation of 15 observations of the matched behavior?
What an interesting study! Another thought for you. It might be interesting to just use Graph Builder to build some box and whisker type plots perhaps with a local data filter embedded, with multiple nested variables across the x axis. Doesn't do much from a formal hypothesis testing point of view but could lend some insight wrt to various hypotheses to propose as well as just exploring the system under study.
Thank you for the information. I will take a look at it.
Do you think it is possible for me to use paired t-test when I only have 1 measurement of the stressor-period?
Or how should I do this? Normalize all data so that I also could use stressor-periods of other chickens to increase my number of measurements of stressor-periods?
There are no labels assigned to this post.