Standardized beta coefficients for xtreg,fe

[Piku Sarkhel]

Dear All,
I am running a panel regression where my independent variable is an index of service quality.
xtreg y Index, fe vce(cluster Country)
Since my independent variable is an index I am trying to obtain the standardized form of beta coefficients.
Can any one kindly suggest how I can do that with xtreg, fe?
Thanking you

[Carlo Lazzaro]

Piku:
you may want to take a look at: st: RE: How to get standardised coefficients running panel data?

[Piku Sarkhel]

Thanks Carlo !!

[Piku Sarkhel]

Carlo Lazzaro the document seems to suggest that I have to standardize the outcome variable and then estimate xtreg. Since my data is a short panel and I intend to estimate a one way country fixed effect model should I standardize the variable over the country id across time ? Would you please suugest ?

[Clyde Schechter]

What is your outcome variable y? Let's think a minute about what standardization does. When you have a variable like an index of service quality, and assuming it is not so widely used that everybody who is in your target audience knows and understands what different scores on it mean, you gain clarity of explanation by standardizing that variable. People are more comfortable with the interpretation of a coefficient denominated in standard deviations of an index than they are with coefficients denominated in index points. (And they may even actually understand them better.) But with a variable whose units of measurement are commonplace, or at least well understood by people in your target audience, standardizing does not improve clarity of explanation. In fact, it makes it worse. It is easy to understand a coefficient whose units numerator is, say, dollars, or meters, or kilograms, etc. But a coefficient whose units numerator is standard deviations of currency, length or mass is completely incomprehensible unless you share the standard deviation with the audience. And even then, the coefficient becomes comprehensible only with strenuous mental arithmetic. So while it sounds like you have good reason to standardize x, whether you should also standardize y depends on what the units of measurement of y are. If familiar to your audience, leave it in its natural units; if arbitrary or unfamiliar, standardization is helpful.

[Piku Sarkhel]

Clyde Schechter thanks a lot for your reply. My outcome variable is death per hundred thousand people . However, the effect size is low . So like I am getting an unit increase in service quality reduces death by 1.6. I was wondering whether this makes sense. I would be grateful if you could suggest if theres any thing is wrong !

[Carlo Lazzaro]

Piku:
why not posting what you typed and what Stata gave you back (as per FAQ)? Thanks.

[Clyde Schechter]

Re #6. It isn't possible to judge that based on the information provided. Are we talking about all cause mortality in some population? Or only mortality from some specific cause, or in some subset of the population? And what kind of services does the index measure the quality of? How directly related are those services to mechanisms of mortality? Does your model also adjust for relevant other variables? It is really difficult to do observational studies of effects on mortality outcomes because there are so many factors that affect it and it is really difficult to design a study that will adequately adjust for that.