Michelle Bell <michelle.bell@yale.edu> 

12/15/2004 11:20 AM

	

To

Jee-Young Kim/RTP/USEPA/US@EPA

cc

Subject

Attachment with ozone/mortality estimates









Young,

I'm resending the attachment. Did it come through this time?

I am sending the betas and stderrs as you requested. Note that all
values 

have been multiplied by 1000 (I find them easier to read, and also is an


excellent approximation of the percent increase in mortality for a 10
ppb 

increase in ozone).

If you divide by 1000, the beta is for 1 ppb increase in daily ozone
(over 

the past week - these are the constrained distributed lag model
estimates).

Michelle

At 08:24 AM 12/15/2004 -0500, you wrote:

>Hello Michelle,

>

>The spreadsheet mentioned below has not been sent along with your

>e-mail.  Would you please e-mail it to us again?

>

>Thanks, Young

>

>

>

> 

>

>Michelle Bell <michelle.bell@yale.edu>    To:  Jee-Young
Kim/RTP/USEPA/US@EPA

>12/14/2004  04:08 PM                      cc:  David
Svendsgaard/RTP/USEPA/US@EPA, Dennis Kotchmar/RTP/USEPA/US@EPA

>                                          Subject:  Re: Figure 2 of
your JAMA paper

>

>

>Hi Young,

>

>Here are answers to your questions.

>

>1) The constrained distributed lag model provides an estimate for the

>exposure over the previous week (same day (lag 0) and lags from 1 to 6

>days). In our model, this is not either structured as either the 7-day

>average exposure or the sum of 7 effect estimates. You can see the
model

>structure in the paper, I believe. If you still have questions, let me
know

>and I'd be glad to answer.

>

>Our unconstrained distributed lag model does sum 7 effect estimates
(one

>for each day of exposure). Ideally we would have liked to look at these

>effect estimates separately to see the different effect by days,
however

>the individual estimates are unstable due to their high collinearity.
The

>sum, however, is stable. You may notice that the constrained and

>unconstrained versions of the distributed lag model provide very
similar

>results.

>

>2) Figure 2 provides the Bayesian (posterior) estimates. I am attaching
a

>spreadsheet that includes the data that was used to generate the
numbers

>for Figure 2. As you know, my colleagues and I are also very interested
in

>the possible heterogeneity across cities.

>

>Please let me know if you have any further questions.

>

>Best regards,

>Michelle

>

>

>At 10:23 AM 12/13/2004 -0500, Kim.Jee-Young@epamail.epa.gov wrote:

> >Hello Michelle,

> >

> >For the current draft of the ozone criteria document, we have been

> >studying all aspects of your ozone-mortality paper published in JAMA.

> >We had one question regarding your constrained distributed lag model.

> >The estimate of 0.52% per 10 ppb increase in previous week's 24-h avg

> >ozone level, does this refer to the effect estimate for the 7-day

> >average exposure, or the sum of the effect estimates for each single
day

> >lag?

> >

> >Also, we would like to examine further the heterogeneity of effects

> >expressed in the 95 communities.  Would you please send us the

> >quantitative data used to plot Figure 2 of the JAMA paper?

> >Specifically, please send us the numerical values of the mean effect

> >estimate and standard error (or preferably, the beta and standard
error)

> >for each of the 95 communities.

> >

> >Thank you.  We look forward to your response,

> >

> >Best,

> >Young

