sampling weights support?

[ignacio martinez]

Hi everyone,

Is it possible to use sampling weights with brms (rstanarm, or stan)? In case is relevant, this is how stata does it if you are using a frequentist approach.

Thanks,

Ignacio

[Paul Buerkner]

Let's say y is your response and w is your weight variable. Then go for

fit <- brm(y | weights(w) ~ <predictors>, ...)

--
You received this message because you are subscribed to the Google Groups "brms-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+unsubscribe@googlegroups.com.
To post to this group, send email to brms-...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/brms-users/a2a5cd9a-3754-4f1f-8144-6e9bdb0ad6d5%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ignacio Martinez]

Thanks Paul. Is the implementation/stan model discussed somewhere?  

On Wed, Sep 27, 2017 at 11:23 AM Paul Buerkner <paul.b...@gmail.com> wrote:

Let's say y is your response and w is your weight variable. Then go for

fit <- brm(y | weights(w) ~ <predictors>, ...)

2017-09-27 17:12 GMT+02:00 ignacio martinez <igna...@gmail.com>:

Hi everyone,

Is it possible to use sampling weights with brms (rstanarm, or stan)? In case is relevant, this is how stata does it if you are using a frequentist approach.

Thanks,

Ignacio

--
You received this message because you are subscribed to the Google Groups "brms-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+...@googlegroups.com.

[Paul Buerkner]

It is discussed in ?brmsformula from a syntactical perspective.

The implementation is just a weighting of the log-likelihood.

2017-09-27 17:28 GMT+02:00 Ignacio Martinez <igna...@gmail.com>:

Thanks Paul. Is the implementation/stan model discussed somewhere?  

On Wed, Sep 27, 2017 at 11:23 AM Paul Buerkner <paul.b...@gmail.com> wrote:

Let's say y is your response and w is your weight variable. Then go for

fit <- brm(y | weights(w) ~ <predictors>, ...)

2017-09-27 17:12 GMT+02:00 ignacio martinez <igna...@gmail.com>:

Hi everyone,

Is it possible to use sampling weights with brms (rstanarm, or stan)? In case is relevant, this is how stata does it if you are using a frequentist approach.

Thanks,

Ignacio

--
You received this message because you are subscribed to the Google Groups "brms-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+unsubscribe@googlegroups.com.

[Paul Buerkner]

You can take a look at the Stan code via the stancode method or directly via the make_stancode function.

[Ignacio Martinez]

Thank you! 

On Wed, Sep 27, 2017 at 11:30 AM Paul Buerkner <paul.b...@gmail.com> wrote:

You can take a look at the Stan code via the stancode method or directly via the make_stancode function.

2017-09-27 17:29 GMT+02:00 Paul Buerkner <paul.b...@gmail.com>:

It is discussed in ?brmsformula from a syntactical perspective.

The implementation is just a weighting of the log-likelihood.

2017-09-27 17:28 GMT+02:00 Ignacio Martinez <igna...@gmail.com>:

Thanks Paul. Is the implementation/stan model discussed somewhere?  

On Wed, Sep 27, 2017 at 11:23 AM Paul Buerkner <paul.b...@gmail.com> wrote:

Let's say y is your response and w is your weight variable. Then go for

fit <- brm(y | weights(w) ~ <predictors>, ...)

2017-09-27 17:12 GMT+02:00 ignacio martinez <igna...@gmail.com>:

Hi everyone,

Is it possible to use sampling weights with brms (rstanarm, or stan)? In case is relevant, this is how stata does it if you are using a frequentist approach.

Thanks,

Ignacio

--
You received this message because you are subscribed to the Google Groups "brms-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+...@googlegroups.com.

[Andrew Mercer]

Was just looking into this. The documentation says that the the log-posterior for each case is multiplied by the weight. That doesn't seem equivalent to multiplying the likelihood by the weight. I may be missing something here though.

On Wednesday, September 27, 2017 at 11:29:24 AM UTC-4, Paul Buerkner wrote:

It is discussed in ?brmsformula from a syntactical perspective.

The implementation is just a weighting of the log-likelihood.

2017-09-27 17:28 GMT+02:00 Ignacio Martinez <igna...@gmail.com>:

Thanks Paul. Is the implementation/stan model discussed somewhere?  

On Wed, Sep 27, 2017 at 11:23 AM Paul Buerkner <paul.b...@gmail.com> wrote:

Let's say y is your response and w is your weight variable. Then go for

fit <- brm(y | weights(w) ~ <predictors>, ...)

2017-09-27 17:12 GMT+02:00 ignacio martinez <igna...@gmail.com>:

Hi everyone,

Is it possible to use sampling weights with brms (rstanarm, or stan)? In case is relevant, this is how stata does it if you are using a frequentist approach.

Thanks,

Ignacio

--
You received this message because you are subscribed to the Google Groups "brms-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+...@googlegroups.com.

[Paul Buerkner]

The difference is just that in bayesian models the prior comes into play as well ;-)

To unsubscribe from this group and stop receiving emails from it, send an email to brms-users+unsubscribe@googlegroups.com.

To post to this group, send email to brms-...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/brms-users/40e9ac61-3242-41b4-9133-8691a8dc3033%40googlegroups.com.