This is a follow on to a previous thread which is now closed.
About ten years ago I was working out the confidence limits for scaled distances.
My collection of photocopied published papers kept from my recent stint back at college contained a number of statistical flaws:
The authors had frequently assumed a "Normal" distribution, this would only be justified when the sample population has 120 or more degrees of freedom (sample size > or equal to 121 measurements). For smaller sample populations a "t" distribution world be required.
Even assuming that a normal population was useable, the authors had also frequently used 1.96*SD as the limit for 95% confidence. This is the limit for a 2 tailled test and describes the central 95%. Applied to the upper end it is defining 97.5% confidence limit!
I think the main culprit was Wilton & Johnston's 1992 quotation (or mis quotation) of Lucole & Dowding (1979).
Keith
About ten years ago I was working out the confidence limits for scaled distances.
My collection of photocopied published papers kept from my recent stint back at college contained a number of statistical flaws:
The authors had frequently assumed a "Normal" distribution, this would only be justified when the sample population has 120 or more degrees of freedom (sample size > or equal to 121 measurements). For smaller sample populations a "t" distribution world be required.
Even assuming that a normal population was useable, the authors had also frequently used 1.96*SD as the limit for 95% confidence. This is the limit for a 2 tailled test and describes the central 95%. Applied to the upper end it is defining 97.5% confidence limit!
I think the main culprit was Wilton & Johnston's 1992 quotation (or mis quotation) of Lucole & Dowding (1979).
Keith