Martin,

Apparently what we are arriving at is: you apply Cox regression to estimate

risk factors determining the event of evacuation (and the time spent in

achieving it), i.e. the time to arrival to ground zero, using as predictors

the floor of origin plus all your other predictors.

Cox regression estimate the impact of risk factors on odds ratios on the

assumption of proportional hazards. This means that any risk factors

increases the odds of the event in a certain fixed proportion, irrespective

of the time elapsed. For instance, if knowledge of the building reduces the

time by a half, the expected reduction (for people of given characteristics

and given floor of origin) would be a half of the baseline or reference

case, regardless of the time elapsed until exiting the building. A

refinement is introducing non proportional hazards, for instance a

time-dependent covariate: one could imagine that time from a certain floor

to the ground is dependent on tiredness, which would be a function of the

number of floors already descended. In other words, the pace of descent

would be decelerating because people become more tired as they go down. If

so, one possible time-dependent covariate may be time itself.

Then for your second problem. If you already know the time taken to descend,

and the floor of origin, you know the descent velocity (minutes/floors),and

this could be used as a predictor of other things like injuries. As you will

be used the observed velocity as predictor, this second analysis is

independent of the previous one, unless you want to use the PREDICTED

velocity instead of using the observed one, but why would you want that?

You sent me this question off list, but I think you would not object my

sharing it and my comments with everybody, would you?

Hector

-----Mensaje original-----

De: Martin Sherman [mailto:

[hidden email]]

Enviado el: Thursday, June 22, 2006 12:09 PM

Para: Hector Maletta

Asunto: Re: Stats Question

Hector: What would you suggest then. There are two things we are trying to

do with the data. First, we want to see what variables are predictive of

rate of evacuation (prior evacuations, knowledge of the building, gender,

age, emergency preparedness, etc). Second we want to see whether rate of

evacuation is predictive of injury during evacuation. Your thoughts much

appreciated. martin

>>> Hector Maletta <

[hidden email]> 06/22/06 10:27 AM >>>

Frederic,

Great idea about using Cox. However, your reasoning applies only to people

that could possibly start descent. All people in or above the floors where

the planes crashed into were not able to do so, and possibly also many or

all the people in the floor/s immediately below, because of the destruction

around that may have killed some of the people instantly, destroyed access

to stairs, or left people otherwise unable to start descending.

The situation is similar to a medical study on survival time after

successful heart surgery. People who died on the operating table without

having their surgery successfully completed should not actually be counted,

I suppose.

Hector Maletta

-----Mensaje original-----

De: SPSSX(r) Discussion [mailto:

[hidden email]] En nombre de

Frederic Villamayor Forcada

Enviado el: Thursday, June 22, 2006 11:08 AM

Para:

[hidden email]
Asunto: Re: Stats Question

Martin:

I see this as a case of censored data. You are underestimating evacuation

time if those who did not survive are not counted. It the towers had

resisted the attack, the mean evacuation time would have been higher than

that computed only with the data from the surviving people. May be Cox's

regression can help you. About the use of the floor number as a covariate,

I think it's a good idea. Any action you take would be speculative. Why

not take a common start time for all the people on tower 1 and another for

tower 2 (the time when the planes crashed), and all data from the people

who did not survive are censored at the time when each tower fell down?

Greetings

Frederic

%%%%%%%%%%%%%%%%%%%%%%%

Frederic Villamayor

Unitat de Bioestadística

Àrea de Desenvolupament Preclínic

CIDF Ferrer Grupo

Juan de Sada, 32

08028 - Barcelona

Espanya

E-mail:

[hidden email]
Tel: +34 935093236

Fax: +34 934112764

WWW: www.ferrergrupo.com

%%%%%%%%%%%%%%%%%%%%%%%

"Sanity is not statistical"

1984 (George Orwell)

Martin Sherman <

[hidden email]>

Enviado por: "SPSSX(r) Discussion" <

[hidden email]>

22/06/2006 15:07

Por favor, responda a

Martin Sherman <

[hidden email]>

Para

[hidden email]
cc

Asunto

[SPSSX-L] Stats Question

Dear List: I have a data set which provides me with the time it took an

individual to evacuate from the WTCs on 9 / 11. (Time they reached

street level minus time they started to evacuate). Currently the

research team is debating how to calculate the rate of evacuation. One

school of thought is to divide the time of evacuation (End time minus

begin time) by the floor the individual started on. The other school of

thought is to use the time itself as the outcome (End Time minus begin

time) and use the floor they started on as control variable (in

regression). I would be interested in how other folks would view this

situation. As always I very much appreciate the time and thoughts of

other individuals. thank you.

martin sherman

Martin F. Sherman, Ph.D.

Professor of Psychology

Director of Masters Education: Thesis Track

Loyola College

222 B Beatty Hall

4501 North Charles Street

Baltimore, MD 21210

410 617-2417 (Office)

410 617-5341 (FAX)

[hidden email]