Obsidian | Level 7

## Can you evaluate Linear contrasts if Model F is not sign.?

For example:

Lamb BW. Model (p-values in parenthesis) is: BW = juniper (0.53)  day (0.02)  juniper x day (0.16).

Considering that juniper is not significant in the model, I don't think I can look at any linear/quadratic trends, correct?

Basically, does the model variable have to be < 0.05 (or 0.1) to evaluate or discuss the linear/quad. trends?

1 ACCEPTED SOLUTION

Accepted Solutions

## Re: Can you evaluate Linear contrasts if Model F is not sign.?

Not necessarily.  The omnibus F test is essentially whether at least one mean is different.  Trend testing is a specific contrast that involves all of the means--I have seen several examples where a trend test is significant while the omnibus F is not.  The single degree of freedom trend test is more powerful because of the assumption of linearity (or quadratic-ness).

This is a different situation than the comparison of two means following the omnibus test, where you already have information that no two means are likely to be different.

Steve Denham

## Re: Can you evaluate Linear contrasts if Model F is not sign.?

Not necessarily.  The omnibus F test is essentially whether at least one mean is different.  Trend testing is a specific contrast that involves all of the means--I have seen several examples where a trend test is significant while the omnibus F is not.  The single degree of freedom trend test is more powerful because of the assumption of linearity (or quadratic-ness).

This is a different situation than the comparison of two means following the omnibus test, where you already have information that no two means are likely to be different.

Steve Denham

Discussion stats