From this statement of yours:
"Econometic methods take one variable at a time to estimate an equation for testing."
That is absolutely wrong. Modern econometric methods estimate the effects of many variables simultaneously.
When deciding on explanations for an economic change it is usual practice to add one variable at a time to the equation.
I do econometrics for a living. This is simply wrong. You put in all the variables that theory would lead you to believe would have an effect. If you want to test for the effect of a new variable, you add it in together with all the other variables that have been previously shown to have an effect.
You don't just throw in everything you can think of into a regression estimation.
No, you add everything that theory and previous empirical work tells you is important.
As anyone who has looked at the classic Lawrence Klein -An Introduction to Econometrics would know the very first equation discussed used one independent variable.
Yes, but that's only there for expository purposes. It makes the algebra simpler, thereby allowing undergraduates to better understand the mathematics behind regression analysis.
In actual practice, no one estimates equations with one independent variable. You won't find more than a handful of empirical papers published during the last 30 years that do such a thing.
And BTW, no modern graduate text begins with a single variable regression; they start with multiple variables right away, using matrix notation.
It is standard to start with one and look for more when its correlation is not sufficiently high. Advanced econometric models use many.
You have not the slightest idea of what you're talking about.
While it has been 30 years since I took an econometrics course
It shows.
the terms you refer me to have no obvious relation to what I was discussing.
You need to look them up.
Tests for spurious correlation are available but that was not at issue.
Okay, let me lay it out for you very simply. You have dependent variable Y and two independent variables X1 and X2. X1 has an effect on Y. X2 has no effect, but it is correlated with X1. If you estimate
Y = a + bX2
and leave out X1 from the regression, you'll find that b is statistically significant, even though X2 has no effect on Y. This is an example of spurious correlation driven by omitted variable bias, i.e. failure to include a relevant variable in your specification.
Autocorrelation can also be detected and accommodated. Even heteroscedasticity.
Well yes, of course, but this has nothing to do with the matter at hand. These issues relate to the distribution of the risiduals. I'm talking about relationships between the test variables.
Perhaps I am not saying what I mean as well I as should. What I meant wrt estimations is that independent variable are examined one at a time to decide for selection within a model. If one is looking at investment one would first look at interest rates then add expected income or some other variable which would explain changes not "explained" by the first two. I am quite aware that a modern model is very complex and has many variables.
I did not mean to imply or state that one variable could explain all or most of the changes in a dependent variable.
There was no need to demonstrate the effect of a spurious variable it is understood. It changes nothing that I said.