Effect Mod/Interaction HW & Simulation App Clarifications

Effect Mod/Interaction HW & Simulation App Clarifications

by June Chan -
Number of replies: 4

Hi All, 

Thank you for engaging in a new and interesting session today, featuring the Simulation App, created by Francois Rerolle. 

As follow-up to some questions raised in class, we are posting a slightly modified HW assignment, that gives further details for q10-12. To (hopefully) make things simpler, we've specified which metrics to focus on for assessing interaction on the additive and multiplicative scales, and clarified that when we ask you to comment on interaction, we mean both scales.  In class, Francois commented that he thought it is important to not only specify "multiplicative" but "multiplicative on the RR or OR scale". This is a fair point.  To simplify things for the HW, we have specified that to assess additive interaction, please use either the Interaction Contrast of the Risk Difference or RERIrr; for multiplicative interaction, let's focus on the RR scale. 

Lastly, a point was raised after class that in EPI203, there is a slide from your effect modification lecture set that mentions that when lines are parallel (with similar legend and set up as the app), this means that there is not interaction on the multiplicative scale.  Please note, the Y axis in that slide is on the log scale, and is different from the Y axis scale on the app. So, this comment regarding parallel lines from epi 203 does not apply for assessing multiplicative interaction visually on the app. 

Thank you and Good luck!

JMC








In reply to June Chan

Re: Effect Mod/Interaction HW & Simulation App Clarifications

by June Chan -
HI All,

A few additional notes on terms used on the App Dashboard and how these relate to terms/definitions from lecture:

The Interaction Contrast (IC) for the Risk Difference is: p11– p10– p01 + p00 Note, this is the same as what is In lecture slides (Part 3, slide 5) and called – "Interaction Contrast on the Additive Scale".

The IC for the risk ratio = RR11/ RR10RR01 = p11p00/ p10p01 NOTE: This is the same as the "Measure of Interaction on the Multiplicative Scale", lecture Part 3, slide 7.

Please note, some refer to the RERIrr as the “interaction contrast ratio”... which is distinct mathematically from the IC for the risk ratio.

Thank you for your patience with the terminology.

Have a good weekend,
JMC
In reply to June Chan

Re: Effect Mod/Interaction HW & Simulation App Clarifications

by Lucia Abascal Miguel -
Hi June,

What do you mean with mirror in Q10b?

Thanks,

Lucía
In reply to Lucia Abascal Miguel

Re: Effect Mod/Interaction HW & Simulation App Clarifications

by June Chan -
Hi Lucia,

I think a better term might be "opposing complement"?

For example, if one response type is "1 1 0 0" going straight across the values for the 2-5th columns in the app, then the "mirror" or "opposing complement" would be "0 0 1 1"
I hope that clarifies.
In reply to June Chan

Re: Effect Mod/Interaction HW & Simulation App Clarifications

by June Chan -
Reminder, as noted in class with Francois during the app demo, on the Interaction Simulation App, the 7th column labelled as 'Interaction' - is indicative of interaction on the Additive scale.