Is it necessary to put “beta” at the back of every multivariate regression command I run?

[Emerald Chang]

Hello again,

Question 1

Was revising my Stata skills by reading a gentle introduction to stata sixth edition recently, to my surprise, as far as I can recall, the appearance of “beta” in this book is more frequent than a few stata books I read back at uni. So I wonder if “beta” is a must? Or it depends on circumstances?

Coz I always run my multivariate/ multivariable regression like this reg y x1 x2 x3 x4 … instead of reg y x1 x2 x3 x4, beta…

One of examples taken from the session 10.1 multiple regression of a gentle introduction to stata sixth edition :

regress env_con educat inc com3 hlthprob epht3,beta

Type env_con as the dependent variable –concern for environmental issues score

Independent variables are:
1) Educat – years of education
2) Inc – annual income in dollars
3) Com3 – identification with the community
4) Hlthprob – health problems
5) Epht 3 – impact of the environment on the person’s own health

Question 2

What is the minium sample size that I should have in order for me to run multivariate regression anlysis in Stata? I was tryingt to run multivariate regression analysis among 3 subjects the other day. In this case, stata only allows me to run univariate regression (reg cytokine i.treatment groups) but not multivariate regression while I was trying to adjust for medical history (reg cytokine i.treatment groups i.medical history) ... what I could see from my multivariate regression results were nothing but plenty of dots ...

Many thanks for the clarification in advance. : ) truly appreciate it!

[Richard Williams]

Q1: I think standardized coefficients are the work of the devil. But some people like them. You don't have to use the beta option, but you can if you find it helpful or if others insist that you provide that information.

You might find the following helpful (you can skip to p. 6 if you are willing to trust the proofs):

https://www3.nd.edu/~rwilliam/stats2/l71.pdf

I like X standardization a bit better than full standardization. It and other options are discussed at

https://www3.nd.edu/~rwilliam/xsoc73994/L04.pdf

Q2: Did you really only have 3 cases??? Did you have repeated measures for them or something? You can't do much of anything if you only have 3 cases! But I suspect you are leaving out important details.

[Emerald Chang]

Many thanks for the reply, Richard .

[Emerald Chang]

yes. I only had three miscarried cases, as my supervisor just wanted to see the difference in cytokine in relation to their stauts, which is miscarried compared to those live birth women. and then adjusted for their medical history 2 miscarriage versus 1 infertile. maybe paired t-test would be the most approtipate analysis but not regression here? But I can't adjust for the thrid variable if paired t-test was used at the first place. Thank you again.

[Nick Cox]

Opinion: The term multivariate regression is best reserved for regressions with multiple outcomes or responses, e.g. as available in mvreg.

[Emerald Chang]

Hi Nick,

Thanks for the reply too. Have a lovely day !