Type 3 Analysis of Effect in Logistic Regression

unnati — Mon, 20 Jul 2020 17:54:08 GMT

In Testing interaction effect for logistic regression. Can someone explain me following output. specially "the Type 3 Analysis of effect". I would like to know that PSA*ethnic interaction is statistically significant or not. If it is significant then why and if not then why not ? The following ouput is created using PROC logistic as below:

As per study lesson it says that PSA*ethnic interaction is statistically significant (p=0.07) but as per my understanding p-value should be less than 0.05 to be consider as statistically significant. May be i am interpreting it wrong.

I appreciate if someone can explain me where i am interpreting it wrong.

proc logistic data=prostate plots(only)=(effect(x=(psa) sliceby=ethnic)
oddsratio (type=horizontalstat));
class ethnic(param=ref ref="black");
model capsule (event="1") = psa ethnic psa*ethnic;
unit psa = 10;
oddsratio 'psa 50 vs 40 for black' psa/at(ethnic="black" psa=40) cl=pl;
oddsratio 'psa 50 vs 40 for white' psa/at(ethnic="white" psa=40) cl=pl;
run;

Re: Type 3 Analysis of Effect in Logistic Regression

Reeza — Mon, 20 Jul 2020 18:00:30 GMT

as per my understanding p-value should be less than 0.05 to be consider as statistically significant.

Your p-value should be less than your cutoff which is typically 0.05 but can also be a different number so it actually depends on exactly how the question was phrased. If it said alpha was 0.1 then the statement it is statistically significant would be correct.

@unnati wrote:

In Testing interaction effect for logistic regression. Can someone explain me following output. specially "the Type 3 Analysis of effect". I would like to know that PSA*ethnic interaction is statistically significant or not. If it is significant then why and if not then why not ? The following ouput is created using PROC logistic as below:

As per study lesson it says that PSA*ethnic interaction is statistically significant (p=0.07) but as per my understanding p-value should be less than 0.05 to be consider as statistically significant. May be i am interpreting it wrong.

I appreciate if someone can explain me where i am interpreting it wrong.
proc logistic data=prostate plots(only)=(effect(x=(psa) sliceby=ethnic)
oddsratio (type=horizontalstat));
class ethnic(param=ref ref="black");
model capsule (event="1") = psa ethnic psa*ethnic;
unit psa = 10;
oddsratio 'psa 50 vs 40 for black' psa/at(ethnic="black" psa=40) cl=pl;
oddsratio 'psa 50 vs 40 for white' psa/at(ethnic="white" psa=40) cl=pl;
run;

topic Type 3 Analysis of Effect in Logistic Regression in Statistical Procedures

Type 3 Analysis of Effect in Logistic Regression

Re: Type 3 Analysis of Effect in Logistic Regression