binary outcomes

[macartan]

Suggest the canonical designers should have binary outcome option using probit type dgp Documentation change to reflect outcome_means etc are defined on the latent space if binary invoked Alernative is to make separate designers.

[jaspercooper]

OK I think for many designers this may not be too hard.

Since we're mostly working in a standard normal space, one approach is simply to do a probit-style transformation of the POs.

That's my preferred approach but we would need to warn (either in function or documentation) that in such cases parameters such as ICC and rho pertain to the latent variable and not the realized Y.

Another more laborious approach would be to have an entirely separate DGP. That would entail either an ifelse logic or an rlang implementation in every one of the designers, which for me has the drawback of making source code hard to read for people that want to write their own designers.

[jaspercooper]

I'm hesitant to create parallel designers because I worry about slippage

[macartan]

SGTM

would it be possible to simply allow and function f that transforms Xb; default is identifiy but people cuolld do logits, probits exponentials, create counts or whatever, at tehri own risk

On Wed, Aug 15, 2018 at 3:14 PM Jasper Cooper [email protected] wrote:

I'm hesitant to create parallel designers because I worry about slippage

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DeclareDesign/DesignLibrary/issues/158#issuecomment-413192846, or mute the thread https://github.com/notifications/unsubscribe-auth/AMJO_RF86Bgsnth6p61G-Vva_-Aw-ntQks5uRB6WgaJpZM4V9yiP .

[nfultz]

Agreed it would better to support a link function - I don't think estimatr has any functionality for that though.

On Wed, Aug 15, 2018 at 7:32 AM macartan [email protected] wrote:

SGTM

would it be possible to simply allow and function f that transforms Xb; default is identifiy but people cuolld do logits, probits exponentials, create counts or whatever, at tehri own risk

On Wed, Aug 15, 2018 at 3:14 PM Jasper Cooper [email protected] wrote:

I'm hesitant to create parallel designers because I worry about slippage

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub < https://github.com/DeclareDesign/DesignLibrary/issues/158#issuecomment-413192846 , or mute the thread < https://github.com/notifications/unsubscribe-auth/AMJO_RF86Bgsnth6p61G-Vva_-Aw-ntQks5uRB6WgaJpZM4V9yiP

.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/DeclareDesign/DesignLibrary/issues/158#issuecomment-413216145, or mute the thread https://github.com/notifications/unsubscribe-auth/AAZjTkMA49Q-7ZeowlSyGGBDnL4y-eR6ks5uRDDygaJpZM4V9yiP .