nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

How to specify contrast coefficients with "Contrast Specification” window？

HXZHAI

Hi there,

I recently bought JMP 15.1.0. Very different to SAS which allows me to use IML procedure to generate the orthogonal contrast coefficients for both equally and unequally spaced treatments, I got this dialogue box to click the + or -. It's easier to construct simple one-on-one comparisons by assigning +1 to one and -1 to another. But what if I want to construct a linear and a quadratic contrast with the doses shown below?

Even with equally spaced treatments, it's very hard to key in values above 2, such as the series of -3 -1 1 3 when v=4 (Orthogonality binds the values in a series: one change changes all). What are the rules for clicking the + or -? Is there a direct way to just key in the coefficient values in JMP?

Best regards,

Accepted answers

Mark_Bailey

Let me confirm the answer by way of an example. I mocked up a similar situation: one categorical factor with four levels. It looks like this:

Then I fit the one-way ANOVA model and test the three contrasts:

The Test Detail report shows the final values for the Linear, Quadratic, and Cubic contrasts. They were achieved by clicking the + or - buttons in the UI. The coefficients from your tables are the number of times to click the button.

NOTE: I just realized that I got the signs reversed for the first contrast (Linear) but it should not change the test result.

All comments

Mark_Bailey

The documentation page for the Effect Details report explains how the interface is used to build contrasts as linear combinations of factor levels.

You might also be interested in Custom Tests of the model parameters.

Both of these features involve linear combinations of LS Means or model parameters, so tests of higher order terms require that the terms are entered in to the model before the OLS fitting.

HXZHAI

Thank you so much, Mark!

I did try both Contrast and Custom Tests which gave me the same results. They're indeed equivalent as you mentioned in another post. I read all relevant posts, very helpful. But still some details to clarify:

1) if we want to click in a series of -5 -3 -1 1 3 5 for linear effect with 6 variables (equally spaced), does that mean I need to click - sign 5 times, - sign 3 times, - 1 time, + 1 time, + 3 times, + 5 times from top to bottom to get the serial coefficients for the contrasts even though they might look different?

2) Custom Tests takes expertise, beyond my knowledge. Using an example, if we have a control, control+1 dose, control + 2 doses, control + 4 doses, control + 8 doses of the same product, how can we construct the custom tests for the linear and quadratic effects of this product?

Best regards,

Mark_Bailey

Let me confirm the answer by way of an example. I mocked up a similar situation: one categorical factor with four levels. It looks like this:

Then I fit the one-way ANOVA model and test the three contrasts:

NOTE: I just realized that I got the signs reversed for the first contrast (Linear) but it should not change the test result.