I am reading a paper that reports "For a 10-fold decrease in x, y increased by 0.50".

The predictor (x) has been log-transformed but no base specified (presumably natural log). There is also the question of what does a 10-fold decrease mean? Is this 100/10 = 10 = a 90% decrease?

Assuming this is the case I am getting confused in trying to work out the manipulation to convert this untransformed effect (10-fold decrease in x) back to a transformed effect (i.e. a 1 unit decrease in log x).

The best I can come up with is: $\frac{0.5}{log(0.1)}$

Any tips would be appreciated.


1 Answer 1


If the details of the regression model are unclear, I recommend you read more deeply into all the relevant materials; if this is still insufficient then you could contact the author to seek clarification. However, given that the author refers to a 10-fold change in the explanatory variable, I think it is more likely that the regression model used a base-10 logarithm, so it was probability a model that looked something like this:

$$Y = \beta_0 + \beta_1 \log_{10} (x) + \cdots + \varepsilon,$$

which gives the corresponding regression equation:

$$\hat{Y} = \hat{\beta}_0 + \hat{\beta}_1 \log_{10} (x) + \cdots.$$

For this model you have:

$$\begin{align} \hat{Y} - \hat{\beta}_1 &= \hat{\beta}_0 + \hat{\beta}_1 [\log_{10} (x) - 1] + \cdots \\[6pt] &= \hat{\beta}_0 + \hat{\beta}_1 [\log_{10} (x) - \log_{10} (10)] + \cdots \\[6pt] &= \hat{\beta}_0 + \hat{\beta}_1 \log_{10} (x/10) + \cdots \\[6pt] \end{align}$$

My guess is that the author got an estimated coefficient $\hat{\beta}_1 = -0.5$, which would mean that a "ten-fold reduction" in $x$ (i.e., dividing $x$ by ten) increases the predicted value of $Y$ by $-\hat{\beta}_1 = +0.5$.


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.