OK, like the nightclub Kenan and Kel founded in Rigby’s shop, we all know this blog is ‘duh bomb’ (or will be as I add more than my solitary post). However, there are other good R blogs. I thoroughly recommend signing up to the R-bloggers mailing list.


Monty Hall Meets Monte Carlo

In 1990, Parade magazine columnist Marilyn vos Savant caused a giant kerfuffle over an answer she gave to a question posed by a reader. The problem, as stated by the omnipotent Wikipedia, goes like this:

Suppose you’re on a game show, and you’re given the choice of three doors: Behind one door is a car; behind the others, goats. You pick a door, say No. 1 [but the door is not opened], and the host, who knows what’s behind the doors, opens another door, say No. 3, which has a goat. He then says to you, “Do you want to pick door No. 2?” Is it to your advantage to switch your choice?

Vos Savant’s answer was you should always switch. The probability of winning the star prize (a brand new Rover Vitesse Fastback) is 2/3 if you switch, and only 1/3 if you stick. However, this answer outraged many people, and 10,000 furious souls dabbed their quills in some ink and wrote strongly worded, and possibly impolite, letters to Vos Savant refuting her analysis. It was Vos Savant who would have the last laugh though, as her answer was indeed correct (LOL).

That people would get this puzzle wrong is understandable. I know I did the first time I was faced with it. The reason is deeply rooted in behavioural psychology. Humans are not good at dealing with conditional probabilities. If only Vos Savant’s irate readers knew how to use Bayes theorem, they could have saved themselves the ink and the emotional turmoil by deriving the analytic solution to the problem (although, on reflection, doing this would have also led to some ink usage).

What if you’re too lazy or unsure how to derive an analytic solution to the problem? Is there a way for you, you feckless badger? Why yes Kent, there is – use simulation. In the below, I have included some simple R code which simulates the problem. As you can see you get close enough to the Vos Savant’s correct answer.

The code below is an example of how it is possible to use simulation techniques to derive solutions when you do not want to, or are too stupid to do the mathematics. The beauty of computer simulation is that it can be used in much more complex settings where analytic solutions are not feasible. Extracting results via computer simulation is crucial in Bayesian statistical analysis, and also in frequentist calculations of model uncertainty like bootstrapping.

# clean
# outcome matrix
cars <- matrix(NA,nrow=10000,ncol=2)
for(i in 1:10000){
   # initial choices
   choices <- 1:3
   cardoor <- sample(choices,1)
   goatdoor <- choices[-cardoor]
   contest <- sample(choices,1)
   choice2 <- c(contest,NA)
   choice2[2] <- ifelse(contest==cardoor,sample(goatdoor,1),cardoor)
   # the guy who sticks
   cars[i,1] <- ifelse(contest==cardoor,1,0)
   # the guy who changes
   cars[i,2] <- ifelse(choice2[choice2!=contest]==cardoor,1,0)