topic Re: How to test difference in two methods? in Statistical Procedures

How to test difference in two methods?

fridge_wpg — Fri, 17 Feb 2017 02:12:03 GMT

Hi,

I have a regression coefficient (y=1.11*x) of a regression model: KM % = 1.11*TEMR. In another independent data set say Set A, I have TEMR readings from the same instrument as well as KM% of 120 samples. Now if in the Set A, I first multiple TEMRx1.11 to get a variable say TEMR_modified and then like to test whether TEMR_modified would be any different from the KM% of the Set A?

What would be my options, please?

I first thought of treating TEMR_modified and Km% (both have N120) as two groups and run a two sample ttest, but don’t know if it would be valid (as I notice some occasional variation in both groups for respective readings).

Please, advise and if possible with a sas code, thank you all

Re: How to test difference in two methods?

PGStats — Fri, 17 Feb 2017 03:29:42 GMT

You can check that your new data supports (does not disagree with) the hypothesis that TEMR = 1.11 KM with proc REG:

proc reg data=A;
model TEMR = KM / noint;
test KM = 1.11;
run;

Re: How to test difference in two methods?

fridge_wpg — Fri, 17 Feb 2017 04:26:09 GMT

Hi,

Thank you very much for your suggestion. I ran the code "test = km 1.11" the particular SAS Output is:

Test 1 Results for Dependent Variable KM Source DF Mean
Square F Value Pr > F Numerator Denominator

1	20.14595	0.33	0.5661
237	61.00256

I guess the default null hypothesis for the test=1.11 would be that the "new regression model's coefficient (TEMR=KM)" is no different than 1.11?

So, what about the question I asked in my original post, how to test that "TEMR_modified is no different than KM%". Or the above test is sufficient? please. Thank you again

Re: How to test difference in two methods?

PGStats — Fri, 17 Feb 2017 04:58:50 GMT

A 2 sample test might not have much power, depending on the range covered by km.

You could run a paired t-test of temr_modified and km. Or simply run proc univariate on diff = temr_modified - km.

If you have the original data that yielded the regression slope estimate of 1.11, you could also test that the original data and the new data yield the same slope.

Re: How to test difference in two methods?

fridge_wpg — Fri, 17 Feb 2017 05:35:46 GMT

Hi, thanks for the reply.

I am tempted to do the proc univariate on the diff (is it the simple subtraction of the two, right?, no squaring of the mean diff etc!!!!).

Your suggestions are appreciated.

Thanks again