1. Econometrics, as defined here, would include anything regression or classification. Indeed, Machine Learning would be included as "Econometrics" by the author's definition.
I think everyone, including econometricians, would say that Machine Learning is doing a pretty decent job at forecasting.
The authors are thinking of classical regression, but maybe this is simply because it's an article from 1978?
The answer may be: Perhaps we just weren't there yet?
2. The primary goal of econometrics aka statistical inference is not forecasting. It is separating effects from confounders, in a multitude of fashions. Modern approaches are bad at forecasting, and this is almost by design. One tries to approximate experiments, in a sense, and experiments by definition do not forecast real, complex situations: They try to get rid of everything that makes reality complex.
3. The definitions survey question can be understood as
- Statistics vs. qualitative research
- Econometrics vs. time-series econometrics
- Economics vs. other social sciences
- Regression vs. other statistical techniques
- ...
In other words, the survey is designed to get the maximum agreement from economists.
The "test", however basically evaluates the success of early simple linear regression studies.
This is, essentially, what we call a biased research/survey design, to the degree that I'd suspect there is some sort of agenda at play!
Please note that this paper is 40 years old and there is no reason for anyone to read it, with the possible exception of someone working on a history project.
1. Econometrics, as defined here, would include anything regression or classification. Indeed, Machine Learning would be included as "Econometrics" by the author's definition. I think everyone, including econometricians, would say that Machine Learning is doing a pretty decent job at forecasting. The authors are thinking of classical regression, but maybe this is simply because it's an article from 1978? The answer may be: Perhaps we just weren't there yet?
2. The primary goal of econometrics aka statistical inference is not forecasting. It is separating effects from confounders, in a multitude of fashions. Modern approaches are bad at forecasting, and this is almost by design. One tries to approximate experiments, in a sense, and experiments by definition do not forecast real, complex situations: They try to get rid of everything that makes reality complex.
3. The definitions survey question can be understood as
- Statistics vs. qualitative research
- Econometrics vs. time-series econometrics
- Economics vs. other social sciences
- Regression vs. other statistical techniques
- ...
In other words, the survey is designed to get the maximum agreement from economists. The "test", however basically evaluates the success of early simple linear regression studies.
This is, essentially, what we call a biased research/survey design, to the degree that I'd suspect there is some sort of agenda at play!