Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Survival analysis and control variables

From   maarten buis <>
Subject   Re: st: Survival analysis and control variables
Date   Wed, 4 May 2011 13:32:16 +0200

--- Wed, May 4, 2011 at 12:15 PM, D.W.Richards wrote:
> My research involves survival analysis to analyse whether a stock is held for longer
> if it is a gain or a loss.  For each stock traded by an investor, I create a time varying
> variable which indicates whether it trades at a loss or not on each day it is held.   I
> then interact demographic variables like age, gender, with the loss variable to
> understand whether the demographic variables reduce the tendency to hold stocks
> at a loss.  For example, age of the investor holding a stock decreases the survival
> time of stocks at a loss.  The equation in a Cox model is:
> h(t)= ho(t) exp ( LossB1 + lossxageB2 + ageB3)
> where ageB3 is a control variable.  I am most interested in B1 and B2 to assess
> whether they influence holding of losses.
> My next step which I am stuck on is adding further control variables.  For example,
> age decreases the selling of losses and investor wealth also decreases the selling
> of losses.  But age is positively correlated with wealth.  I want to discover the effect
> of age after controlling for wealth.
> How do I go about doing this?

Just add wealth to your model. So estimate:

 h(t)= ho(t) exp ( LossB1 + lossxageB2 + ageB3 + wealthB4)

The real question is: what does age mean nett of wealth and other
control variables? I
find it hard to believe that age has a direct effect. I would think
that age primarily works
through a set of indirect effect, you mentioned wealth, others are
position in the labor
market, risk aversion and time preference. If you start controlling
for these you will just
end up with random noise. Remember, we only want to control for confounding
variables but _not_ for intervening variables.

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index