12-01-2014, 02:26 PM

Hi All,

I have a sample of observations X() and calculate the following:

1) MinX is the minimum value in X().

2) MaxX is the maximum value in X()

3) N is the number of observations.

4) LowCI is the lower confidence interval for the observations X()

5) UpCI is the upper confidence interval for the observations X()

Do you see any value in the following statistics

1) (MaxX - MinX) / (UpCI - LowCI)

2) [(MaxX - MinX)/N] / (UpCI - LowCI)

I can see that the above ratios give an indication of the ratio of the range defined by the extreme values divided by the confidence range as calculated using the entire set of observations. High ratio values may indicate data dispersion and vice versa.

What if we make the following changes:

1) MaxX is the mean value of the highest n values in X().

2) MinX is the mean value of the lowest n values in X().

By taking the mean of a few extreme observations (say 2 or 3 in small samples) we avoid the outliers Max(X()) and Min(X()).

Any other insight or suggestions?

Thanks,

Namir

I have a sample of observations X() and calculate the following:

1) MinX is the minimum value in X().

2) MaxX is the maximum value in X()

3) N is the number of observations.

4) LowCI is the lower confidence interval for the observations X()

5) UpCI is the upper confidence interval for the observations X()

Do you see any value in the following statistics

1) (MaxX - MinX) / (UpCI - LowCI)

2) [(MaxX - MinX)/N] / (UpCI - LowCI)

I can see that the above ratios give an indication of the ratio of the range defined by the extreme values divided by the confidence range as calculated using the entire set of observations. High ratio values may indicate data dispersion and vice versa.

What if we make the following changes:

1) MaxX is the mean value of the highest n values in X().

2) MinX is the mean value of the lowest n values in X().

By taking the mean of a few extreme observations (say 2 or 3 in small samples) we avoid the outliers Max(X()) and Min(X()).

Any other insight or suggestions?

Thanks,

Namir