0

When studying likelihoods, I was presented with the following example:

A drug is given to $n$ patients. We observe $y_1, \dots, y_n$ where $y_i = 1$ if the $i$th patient is cured and $0$ otherwise. Our model may be that $y_i$ is a realisation of $Y_i$, and $Y_i$ has a binomial distribution with parameter $p$; and that the $Y_i$ are independent of each other.

The solution is provided as follows:

Assume $n = 20$ and $\sum_{i = 1}^n y_i = 12$. We give the plot of $x^{\sum_{i = 1}^n y_i}(1 - x)^{n - \sum_{i = 1}^n y_i}$

The R code provided is as follows:

n <- 20
sumy <- 12
curve(x^sumy * (1 - x)^(n - sumy), from = 0, to = 1, n = 301, xlab = "p", ylab = "L(p; y)")

The binomial PMF is $$Pr(k; n, p) = Pr(X = k) = {n \\ k \choose} p^k (1 - p)^{n - k}.$$

There seems to be a discrepancy between these two, where the ${n \\ k \choose}$ term is absent in the example solution binomial PMF. Why is the ${n \\ k \choose}$ term absent in the example solution binomial PMF?

The Pointer
  • 1,064
  • 13
  • 35

1 Answers1

0

Presumably, this code is meant to show that the maximum likelihood estimator is 12/20. To answer your question, the term ${n \choose k}$ does not affect the maximization of the function since it only modifies the likelihood by a constant of proportionality that does not depend on the unknown parameter. Thus, the point on the x-axis where the maximum occurs is unchanged by this constant. To see it analytically, differentiate the log-likelihood with respect to $p$ and you will see that $\log{n \choose k}$ drops out.

hard2fathom
  • 520
  • 1
  • 3
  • 7
  • Thanks for the answer. So $x^{\sum_{i = 1}^n y_i}(1 - x)^{n - \sum_{i = 1}^n y_i}$ is the likelihood? – The Pointer Feb 17 '20 at 04:50
  • 1
    The likelihood is $P(X=k) = {n \choose k}p^k(1-p)^{n-k}$ for given $n$ and $k$, where we view it as a function of $p$. Sometimes, some people omit the term ${n \choose k}$ since it does not factor into calculation of the MLE. – hard2fathom Feb 17 '20 at 04:52
  • Ok. Thank you for the clarification. – The Pointer Feb 17 '20 at 04:53