From 71c6b8d867f97af6ead5268507c7cbf71d2540f7 Mon Sep 17 00:00:00 2001 From: Christian Lorentzen Date: Fri, 8 Dec 2023 19:37:40 +0100 Subject: [PATCH] DOC fix text on null hypothesis in quantile example (#126) --- docs/examples/quantile_regression.ipynb | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/docs/examples/quantile_regression.ipynb b/docs/examples/quantile_regression.ipynb index 807363e..96cad12 100644 --- a/docs/examples/quantile_regression.ipynb +++ b/docs/examples/quantile_regression.ipynb @@ -451,7 +451,8 @@ "metadata": {}, "source": [ "The test statistic is negative which means that the score of the linear model is better than the one of the true quantile.\n", - "The very large p-value, however, leaves us no room but to reject the null hypothesis of equal mean values, i.e. equal scores.\n", + "With the very large p-value, however, we cannot reject the null hypothesis of equal mean values, i.e. equal scores.\n", + "It means that we do not have statistical evidence to say that any one of the linear and the true model performs better.\n", "\n", "We play the same game and perform a t-test for the score difference of the linear and the HGBT models." ] @@ -488,7 +489,7 @@ "source": [ "This time, the p-value is 26%:\n", "Assuming the null hypothesis of equal scores is true, there is a 26% chance of the test statistic to be more extreme as our obtained value of -1.126.\n", - "This is not splendid, but given our sample size it is small enough to say that the linear model has a significantly better score than the HGBT model.\n", + "This is not splendid, but, given our small sample size of 100, it is small enough to reject the null hypothesis and say that the linear model has a significantly better score than the HGBT model.\n", "\n", "All of the above score analysis was based on the pinball loss, but there are other possible choices.\n", "Does the ranking of the models change with the choice of a scoring function?\n",