5
$\begingroup$

I would like to calculate an estimate (even a very rough one if it is the best I could get) of the relative importance of predictors in a logistic regression, something which can let me tell a common person, not proficient in statistics (just like me), for example: these are the predictors: x1, x2, x3, x4, They are all significant statistically, but as you can see, x2 is more important than x1, x3 and x4, because the value of "unknown datum" is higher than "unknown value/other predictor".

I read Relative importance of predictors in the final model, importance of each predictor in logistic regression, How to quantify the Relative Variable Importance in Logistic Regression in terms of p?, but I did not read what I am looking for. I use R and I know a package for it: caret (https://github.com/topepo/caret/) and its function called 'varImp', but I can not understand how the absolute value of t-statistic is used, so I can not understand how to comment values obtained through this formula.

  • The main question is: how can I tell, in practice, that one predictor of a logistic regression is more important than another?

  • Secondarily: how can I tell how much a predictor in that logistic regression is more important than another? Can you explain to me, a statistician wannabe, how the absolute value of t-statistic could be helpful?

$\endgroup$
11
  • 3
    $\begingroup$ There is no agreed upon importance measure for predictors in a model, and the t statistic is not meant to be used in this way, the use of it in this way in carat is a dubious decision. Indeed, variable importance is just a slippery and borderline undefined concept. So heres a question: if I told you a measure, what would you use it for? What kinds of decisions would you like to make or actions would you like to take as a consequence of this measurement? $\endgroup$ Jul 29, 2017 at 18:10
  • 1
    $\begingroup$ Thank you for your answer. No decision would be made basing on those results: it would be just a way to tell someone how much a predictor is more relevant than another one in a specific logistic regression. I am thinking about something shaped like variable inflation factors, that is, simply a number which is somewhat interpretable $\endgroup$ Jul 29, 2017 at 21:23
  • 3
    $\begingroup$ I'm not the downvote, but I will say that if you cannot point to any specific and detailed reason for wanting to know such a thing, then this question is unanswerable. "Predictor Importance" is just not a statistically meaningful thing, there is nothing the universe gives us that you are trying to estimate. The universe tends to mix all it's effects together, and there is rarely, in any real situation, a meaningful way to decompile their effect on some other phenomena... $\endgroup$ Jul 29, 2017 at 22:08
  • 1
    $\begingroup$ This means that concepts like that can only be made sense of within a given context: I.e., I can take an action that looks like this, what is the most beneficial place to take this action to achieve my goals? If you can not provide such a context, then I'm left questioning what you are really after. $\endgroup$ Jul 29, 2017 at 22:08
  • 3
    $\begingroup$ @statisticianwannabe Ok, so in that example, what does it mean to be "more relevant", and how would such a measure inform a consumer of the model? I don't see any situation where saying age is more or less relevant than blood pressure solves a meaningful problem. For example, people can not control their age, but they have some meaningful control over their blood pressure. So a doctor can't say, "well your blood pressure is high, but research has shown age is a more important factor, so you should work on lowering that first." Sor really, what problem are we trying to solve? $\endgroup$ Jul 30, 2017 at 17:22

1 Answer 1

7
$\begingroup$

I assume all predictors have been standardized (thus, centered and scaled by the sample standard deviations).

Let $\mathbf{x}$ be the vector of predictors and $y$ the response, conditionally Bernoulli-distributed wrt $\mathbf{x}$. Then if $\mu=\mathbb{E[y|\mathbf{x}]}=p(y=1|\mathbf{x})$, then clearly

$$\frac{\partial \mu}{\partial x_i}=\beta_i \frac{\exp{(-\beta_0-\boldsymbol{\beta}^T\cdot \mathbf{x})}}{(1+\exp{(-\beta_0-\boldsymbol{\beta}^T\cdot \mathbf{x})})^2}$$

measures the effect of $x_i$ on $\mu$. This effect is a function of $\mathbf{x}$. However, the relative importance of two predictors is

$$\frac{\frac{\partial \mu}{\partial x_i}}{\frac{\partial \mu}{\partial x_j}}=\frac{\beta_i}{\beta_j}$$

which is independent of $\mathbf{x}$. Thus, provided we have standardized all predictors, we can look at the estimates of the model coefficients as indicators of the relative importance of the predictors for what it concerns the variation of the output.

As an example application, I will adapt the case in section 4.3.4 of An Introduction to Statistical Learning, by James, Witten, Hastie & Tibshirani. Suppose you have a database Default of default rates for credit card owners, with predictors student (categorical), income and credit card balance (continuous). Standardize the predictors and fit a logistic regression model. Now you can use the relative magnitude of the $\hat{\beta}_j$to decide which predictor has a larger effect on probability of default. This helps the credit card company decide to whom they should offer credit, which categories are the more risky, which customer segment to target with an ad campaign and so on.

Finally, this paper lists six different definitions of relative predictor importance for logistic regression.

The first one is very similar to the one I showed, with the only difference that instead of standardizing the predictors before, they standardize the $\hat{\beta}_j$ after estimation by multiplying by the ratio $\frac{s_j}{s_y}$ where $s_y$ is the response sample standard deviation, and $s_j$ is the sample standard deviation of predictor $x_j$. It's not exactly the same as my suggestion, because the estimators for the logistic regression coefficients are nonlinear functions of the data, but the idea is similar.

The second one (using the $p-$values from the Wald $\chi^2$ test) is flawed, as explained by @MatthewDrury in the comments to the OP, and shouldn't be used.

Third one (logistic pseudo partial correlation) can be a good choice as long , instead of the Wald $\chi^2$ statistic, in the numerator of the pseudo partial correlation we use the ratio of the likelihood of the model with just predictor $x_i$, to that of the full model. I cannot comment on the other approaches since I don't know enough about them.

$\endgroup$
4
  • 2
    $\begingroup$ +1. I think your answer would be improved with a short case study of how this formula could be used in scientific or business decision making. Like the one in the comments above! $\endgroup$ Jul 30, 2017 at 21:37
  • 1
    $\begingroup$ Will do. So I can also properly acknowledge the source for that example :) I thought I already did that, but comments are not as easy to edit as answers, and the link was lost. $\endgroup$
    – DeltaIV
    Jul 31, 2017 at 7:14
  • 1
    $\begingroup$ Thank you for your answer, very interesting and it gave me a clearer vision of the issue. Now, of course, it is a matter of studying more its theoretical bases $\endgroup$ Jul 31, 2017 at 20:04
  • $\begingroup$ Can you please elaborate why the predictos must be standardized beforhand? I guess it is to make their measurement unit dimensionless, but wonder whether this particular scaling is appropriate for all distributions of predictors (provided, they are considered as random variables). Moreover, can you please explain how to standardize categorial predictors? $\endgroup$
    – cdalitz
    Dec 22, 2021 at 15:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.