..even if our design is ex-post facto, non-experimental?
I have observed in many papers where researchers make recommendations of intervention (e.g. increase salary of employees to improve their performance) on the basis of strong positive relationships arising from non-experiments.
@Ed, I have also observed the same. But we are always told that strong correlation is not the same as causality. But, in experimental research that may be carefully duplicated, will the same strong correlation at all times, in all places, allow the author/s to suggest causality? What about this scenario?
Probably an example helps:
The fact that one celebrates his 70s birthday is perfectly correlated with not dying in a car accident at the age of 20.
Obviousely, neither the 70s birthday-party causes to have no accident, nor does the absence of a deadly accident at an age of 20 cause the birthday party.
This example shows that
a) one needs a reasonable causative MODEL to think about a possible cuasation at all,
b) the time of events and the direction of the possible causation must be considered,
c) other possible confounders must be (logically or -better- experimentally) excluded.
Simple correlational studies often lack all three of these requirements. Often there is no concrete causative model proposed, further is is impossible to see in what time-order the events happen, and there are usually more than a million possible confounders.
A clear conclusion about causation can only be derived from a planned experimental study where the suspected cause is controlled by the experimenter, everything else is kept constant, and the experiment is designed to read the response in time after the suspected cause could operate (according to a well described model of how this may happen - the causative model).
No, certainly no. Causality is not identical to correlation and there is a great debate on this. Actually there is not even one broadly accepted causality test, since Granger causality is not the general case and works only for time series data.
There are many papers and books (e.g., see Pearl, Cook & Campbell and Shadish, Angrist, Rubin, Holland, etc.) talking about use of non-experimental data for exploring causal models. It is not simple and requires assuming a lot about the causal network involving the variables (though as Pearl points out, assuming causality from an experiment also requires assumptions).
It is important to realize that causal and associative statements are different. Sometimes people are more interested in one, sometimes people are more interested in the other (one type is not inherently better than the other).
Article Causal and associative hypotheses in psychology: Examples fr...
Statistical methods help us in decision making, and however the reliability of the results we get should be handled with due diligence. There is no commonly accepted methodology in data analysis, and hence this should guide us in interpretation of results. Also elimination of some variables causes results to come other way. However, there is no problem of interpreting the results of a best model in an economic sense. Statistical methods are very crucial, we do not rule them completely. There is need to undertake appropriate data tests to obtain best results.
Dear @Eddie, I think that the statement becomes true if we add an additional condition to become as follows:
A strong, significant relationship between variables is sufficient to assume causality under the condition of not neglecting other affecting factors.
Dear @Eddie
It is only true once you have support based on your theoretical understanding. Otherwise causality can not be estimated through correlation.
It seems that Bradford-Hill criteria (http://en.wikipedia.org/wiki/Bradford_Hill_criteria) are mostly focused on and used in health sciences, especially epidemiology. I'm curious about the criteria's potential use in social sciences (problems, issues, modifications). In particular, I'm interested in causality criteria, used in information systems discipline (examples and references are welcome).
It has been said that statistical significance is a little like being pregnant--ether you are or you aren't. The most important aspect of determining causality is the design of the experiment. If the design includes true randomization and controls the internal threats to validity, then statistical significance is suggestive of causality. If it is significant, you can then get a better idea of the impact by examining the practical significance of the outcome by estimating a measure of magnitude of its impact.
- Badges
- Science topic