JGardner

August 29th, 2009, 08:20 AM

Hi - In 1986 I bought a TI-74, and I'm still not done with it...

I'm working on a program to extract various statistical indices from

a table of data.

Data occurs in the range (40 < n < 450). Data outside

this range is not accepted by the routine. The vast majority of the data will

occur in the range (80 < n < 250).

The value is manipulated thus... n = INT( n / 1.8 + .5 ) and the result

stored as one byte.

INT(expression) returns the the largest integer less than or equal to

the expression.

When the value is retrieved for summation, mean, std deviation, and so

on the value is pre-processed as n = INT( n * 1.8 + .5 ).

My experience with this scheme is that the value returned is always

within +/- 1 of the original value, with the variation seemingly evenly

distributed.

Data is evaluated in subsets, ranging from ~100 to 600 values, typically.

Individual datum vary in accuracy as follows...

90% of the data are within 10% of the actual value. All of the results are

within 20% of the actual value.

My sense is that the data compression scheme is not materially affecting

the results, but I don't know how to address this rigorously, so I'd appreciate

any input - Or even better, a pointer to how to evaluate the scheme.

thanks, Jack

I'm working on a program to extract various statistical indices from

a table of data.

Data occurs in the range (40 < n < 450). Data outside

this range is not accepted by the routine. The vast majority of the data will

occur in the range (80 < n < 250).

The value is manipulated thus... n = INT( n / 1.8 + .5 ) and the result

stored as one byte.

INT(expression) returns the the largest integer less than or equal to

the expression.

When the value is retrieved for summation, mean, std deviation, and so

on the value is pre-processed as n = INT( n * 1.8 + .5 ).

My experience with this scheme is that the value returned is always

within +/- 1 of the original value, with the variation seemingly evenly

distributed.

Data is evaluated in subsets, ranging from ~100 to 600 values, typically.

Individual datum vary in accuracy as follows...

90% of the data are within 10% of the actual value. All of the results are

within 20% of the actual value.

My sense is that the data compression scheme is not materially affecting

the results, but I don't know how to address this rigorously, so I'd appreciate

any input - Or even better, a pointer to how to evaluate the scheme.

thanks, Jack