3
$\begingroup$

Glen_b and I have been debating how to edit this particular question. My view is that a title specific to the contents of the question is necessary as a summary of the question contents. The title, as I edited it, allows one to quickly understand that the question is about using trig functions in regression without venturing further than the "headline."

On the other hand, Glen_b's title is less descriptive of the question body than mine. Its generality may have merit, though, as it can include several types of transformations for regression analysis under its aegis. Rather than face an avalanche of questions asking "how do I apply __ transform in my regression," we might use this question to capture many transformations at once.

I do not think that Glen_b's approach is desirable.

  1. Imprecise question titles are a pet peeve of mine, as it requires one to read the question body to discern whether the result is relevant to what I'm after. This peeve may not have wide appeal, in which case I'll accept that I'm an outlier.
  2. Collecting all transformations used in regressions into a single question, which is itself about just trig, seems like a very larger task, especially if we admit every transformation satisfying the title "Linear regression with transformed $x$ values." This would appear to include $\log(x)$ and $\log(x+c)$ for some small $c$ (since we need to account for $x$ values which may be 0), $\sqrt{x}$, polynomials, and splines. I'm not sure whether or not the title admits things like Gaussian processes.
  3. While I appreciate that Glen_b is forward-looking in his attempt to service future users, I'm not sure that expanding OP's question does a great service to OP. OP asked about one specific transformation and may be overwhelmed by a very expansive response.

(I hope that Glen_b finds my summary of his position fair & accurate; he and other CV members are, of course, welcome to make their own cases one way or the other -- my intention in describing his argument is merely expository.)

$\endgroup$
  • $\begingroup$ Well, I'd have put things a bit differently, but there's no problem, since I can explain myself in an answer (and have). Thanks. I'd upvote your question as a good question to raise, but this is more of a "choose between approaches" type of question, $\endgroup$ – Glen_b Jul 29 '15 at 23:42
  • $\begingroup$ By the look of things, this appears unlikely to be resolved any time soon -- if at all. It seems unlikely that there will be clear support for the generalization of the title but on the other hand the newer title is not a very substantial change from the original title (that is, if we aren't generalizing the title, the OP's original choice - which is at leats reasonable - should be respected). As a result I think we should be conservative, so I'm reverting to the OP's original title until we have some clearer sense of there being a reason to choose something else. $\endgroup$ – Glen_b Jul 30 '15 at 20:32
  • 1
    $\begingroup$ @Glen_b fair enough. Perhaps the solution is simply to create and self-answer a general question of this variety. $\endgroup$ – Reinstate Monica Jul 30 '15 at 20:39
3
$\begingroup$

I'd better explain my actions and position.

  1. I answered the OP's question in the specific (i.e with direct reference to the details of their question)

  2. I made the title more general --

    changing "Linear regression with sine/cosine elements" to "Linear regression with transformed x values" and answered the more general version of the question as well.

    (general titles with specific bodies are common here; the aim of the more general title was to get hits from users searching for things like "transformed x", while deliberately leaving the question body concrete for the mathematically weaker reader who tend to find it easier to go from the concrete to a similar concrete problem; I think the mathematically able reader who finds it easy to go from the general to the specific would probably find the solution obvious in any case)

  3. The ultimate point (beside answering OP's specific problem and redirecting some kinds of searches) was to provide a question/answer that other specific "how do I deal with this particular transformed-x" type questions could be closed-as-duplicates of.

    There's an older question in similar style (but different specifics) that I point to in my answer, but I thought the new question was more easily generalized, and that's why I worked on it.

  4. I think having an ever-increasing host of specific "how do I fit this transformed-x" questions is counterproductive. Once this issue is resolved, the older one might even be closable, but I'm not overly concerned about it either way (e.g. if it stayed open but with a link to the new one that would also do), as long as we don't end up with dozens of specific versions of this style of question.

So I want to revert the title, but I wasn't going to have an edit war; as such I have for the present left the title as user777 changed it.

[Clearly user777 is trying to do the best for the site; we just don't see eye to eye about the right way to proceed here. I support the choice to post the question to meta.]

$\endgroup$
1
$\begingroup$

What is the proper balance between question & title specificity?

In this answer, there is some guidance about writing titles. The item number 2 says:

Keep the title short: it's not necessary to explain the question in the title, but what it is about without being too broad.

A key part is 'the title should express what the question is about'.

The original title of that question was:

Linear regression with sine/cosine elements

It does not reflect what the question is about, because the connection between 'linear regression' and 'sine/cosine variables' is empty.

The suggestions of Glen_b and user777 add the term 'transformation' to improve the connection between the method (linear regression) and the data (sine/cosine transformed variables):

Glen_b suggested:

Linear regression with transformed $x$ values

user777 suggested:

Linear regression with x values under trigonometric transformation

Both are better than the original title. Note that the four points made in Glen_b's answer here would still be applicable under user777's title suggestion:

  1. The answer answers the original poster's question.
  2. The general part of the answer emphasizes the sine/cosine transformations are irrelevant to the specific issue raised by the OP.
  3. Other questions with the same reasoning, but with different examples of transformation could still be marked as duplicates.
  4. 3 won't let 4 happen.
$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .