1
$\begingroup$

I am new to multivariate analysis and without a great math background I have been able to follow the book Applied multivariate statistical analysis up until this section...

enter image description here

The way I am able to follow the book is mostly through the graphical aids; the math part is not easy to understand for me... I picture bivariate normal, as weight and height of (ie 1000 people)... I can picture in my head the density (with clear correlation)...

My question is: per the book if X1 or X2 is given, the density changes... how come the density changes? When I picture height and weight as a bivariate density, there is no way the density could change, I see both height and weight as given.

(I do understand the 'given' concept in probability, but apparently not here...)

$\endgroup$
0
$\begingroup$

I was going to plot it myself, but looking online, this page does such a good job of it already. They even let you zoom in and swivel the graph. Look at the 2nd from the last graph, the 3d one. So if p(y|x=4), first go to x=4 on the x axis and then look across all the values of y, the z axis now tells you the probability density of y given x=4. Across all the y's, what is the range of the z axis (the density of the probability), not very large.

Looks like they get a maximum probability density near y=0 of approximately 0.005 at x=4. That is, p(y=0|x=4)=0.005 (approximately).

Now look at it when x=0. now the probabilities for y are much higher. p(y=0|x=0) is around 0.03.

Without being conditional on x, y is still a normal distribution, that is, it'll still be that bell shaped curve, but the precise nature of that bell shaped curve changes a lot once you condition it on x. In graphical terms, in this example, it flattens y's curve out if the x value is far away from 0 (in the tails).

$\endgroup$
  • $\begingroup$ So, with my height, weight example would be, a function of height, given weight... For example (height|weight=80kg)? And this would still be normal? $\endgroup$ – Chicago1988 Dec 18 '19 at 15:52
  • $\begingroup$ Yes, f(X_1|X_2) is normal if X_1 and X_2 are normal. For other combinations of distributions, you use the formula above in order to calculate that particular conditional distribution. $\endgroup$ – Huy Pham Dec 18 '19 at 16:12
  • $\begingroup$ Just to help clarify more (hopefully it clarifies instead of confuses). The function f tells you the probability of any y given any x. You could sub in any particular x value then it would tell you the probability of any y given that particular x. Subbing in a particular y as well, then tells you the probability of that y given that particular x. Sub in y without subbing in x and it'll still be a function; a function that describes the prob of y in terms of possible x values (i.e. given x) $\endgroup$ – Huy Pham Dec 18 '19 at 16:17

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.