UP | HOME

p-value

1. as relates to linear regression

1.1. statistical model

  • \(Y=\beta X + \epsilon\) where \(\epsilon\) is the noise which is 0 mean \(\sigma\) variance.
  • In a regression problem, we have data \((x_i, y_i)\) which we say is sampled independently from each other
  • If we consider the \(x\) 's to be fixed, then \(y\) is a random variable
  • And the line of best fit \(\hat{\beta}\) is also a random variable
  • And in fact \(\mathbb{E}[\hat{beta}] = \beta\) where \(\beta\) is the true line of fit (which we will never observe, but assume to exist)
  • And if we assume that \(\epsilon\) is Gaussian, we can get the distribution of \(\hat{\beta}\), if we know \(\beta\)
  • So, the p-value of \(\hat{beta}\) is the chance that \(\hat{\beta}\) takes the observed value if we assume that \(\beta=0\)
  • extremely useful slides
  • notes
  • glm

2. see also

Created: 2024-07-15 Mon 01:28