I've been wondering how the bootstrap method works in the case of nonlinear regression models where the order of the x-axis (suppose just 1 explanatory variable) matters.

In the case of univariate numerical data, as far as I know, the order of the bootstrapped samples is not important because, usually, a summary statistic is computed, but with bivariate data where one variable represents a rank, I'm not sure how it works or if it's correct to use bootstrap.

For example, consider the following model: $$\mathrm d(\log_{10}k)=\frac{1}{\sigma \sqrt{2\pi}}\int_{-\infty}^{\log_{10}k} e^{-\frac{(y-\mu)^2}{2\sigma^2}} \mathrm dy,\quad k \in\mathbb{N}$$

Here is a plot using real observed data: enter image description here

And using a bootstrapped sample:

enter image description here

In both plots, the x-axis is $\log_{10}k$. I suppose that in the latter case, trying to fit the model wouldn't work.

Contrary to my intuition, when I use curve_fit from scipy in bootstrapped samples, I get similar estimated parameters to those I get with the original sample.

This plot shows instances of the model using the original estimated (the smooth almost indistinguishable red line) and bootstrapped (in blue) parameters:

enter image description here

Is curve_fit implicitly reordering the x-axis before fitting? If that were true, wouldn't it affect the bootstrapped method?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.