I think the basic analysis plan should probably be something what I described, the details are up for debate, though. Getting a bit more complicated, I suspect that the best option would be to make a linear mixed-effects model with Time as a fixed predictor and with by-subject random intercepts and slopes for Time; then you would get a confidence interval for the fixed effect of Time (or you could get every participant's individual linear fit by adding the BLUPs [best linear unbiased predictors] to the fixed effect, and then get a confidence interval of those linear fits) and see if the particular subject's linear fit is outside of that confidence interval. That's still the same basic logic as what I suggested above, just a different implementation.

Comparing one subject to the group should be the same as comparing the group to one subject. It's possible, just depends on the method you are using and the nature of your variables. For example, a one-sample t-test can be against any number, so if your dependent variable is continuous you can use a one-sample t-test to compare the mean of the rest of the group against the observed value of the particular subject.

Hello Juliana.  You say you want to compare "the fluctuation on one variable of only one subject to the fluctuation of the same variable of the rest of the group" (emphasis added).  Does this mean you have multiple (repeated) measures per person over time?  If so, are the time points fixed (and the same for everyone) or variable?  Have you looked at a scatter-plot?  Does it look like a linear fit would work well, or does it look like some more complicated model would be required?  HTH.

I agree with Stephen

Thank you all for your contributions.

Regarding to the design, in fact, I have repeated measures per person over time and the time points are fixed and the same for every single participant in the sample. I think a linear fit would work... I want to know if one subject fluctuation is significantly different to the mean of the fluctuation of the entire group in this time lapses.

Could you explain a little better what a scatter plot is?

What else would you suggest??

If you Google "scatter plot" you'll know what it is. Like Bruce suggested, if you make a scatter plot you'll be able to see how this subject compares to the others.

As for significance testing, you can make a difference score (i.e., timepoint 2 minus timepoint 1) or linear fit for every subject. Then you can get the confidence interval of the other subjects, and see if this subject is inside or outside of that confidence interval.

Thank you for your help, for real.

Now I'll look for these analysis in spanish, so I can understand better which are the best suited for my research.

Thank u again.

I think the basic analysis plan should probably be something what I described, the details are up for debate, though. Getting a bit more complicated, I suspect that the best option would be to make a linear mixed-effects model with Time as a fixed predictor and with by-subject random intercepts and slopes for Time; then you would get a confidence interval for the fixed effect of Time (or you could get every participant's individual linear fit by adding the BLUPs [best linear unbiased predictors] to the fixed effect, and then get a confidence interval of those linear fits) and see if the particular subject's linear fit is outside of that confidence interval. That's still the same basic logic as what I suggested above, just a different implementation.

Stephen is absolutely right here, a mixed effect model would be the most comprehensive answer to your question.

Hi all,

nice discussion. From everyday medical practice we know that we are using standards. We take a blood sample or examine ecgs. When the test is done we compare the result with what is "normal". In blood test we mostly have a margin of results, don't know the English term (Normbereich in German) so if the result is below or above this margin for normal results, we take into consideration if this finding is pathological. As for heart rates: we detect bradycardy and tachycardy. 

So regardless of what your research is exactly about. If you manage to establish which is normal you can compare an individual record with what is normal without further statistics. You can just state, we have found that the normal record shows result between these two (lower and upper limit) records, the record of this individual is not normal.

Sorry for my bad English in this regard, I write English a lot but not in this kind of context.

Hope this helps.