Actuarial Outpost
 
Go Back   Actuarial Outpost > Cyberchat > Non-Actuarial Topics
FlashChat Actuarial Discussion Preliminary Exams CAS/SOA Exams Cyberchat Around the World Suggestions

Registration Form
Be Notified

of New
Actuarial Jobs

D.W. Simpson
& Company

International
Actuarial
Recruiters

www.dwsimpson.com

Australia Jobs
Casualty, Life,
Superannuation,
Health &
Investment

Pension
Actuarial Jobs

Consulting &
Insurers --
Students,
Associates &

Fellows


Reply
 
Thread Tools Display Modes
  #1  
Old 02-05-2011, 12:51 PM
UFActuary UFActuary is offline
Member
 
Join Date: Jul 2005
Posts: 2,015
Default Is 50th percentile of data, the mean? or median?

What I don't understand is if the median and mean are not the same (there's a skew).....in a sample

Then the amount by which your score is higher than half the sample, would be the median, NOT the mean.

So then if you calculate percentile by taking x - mu/ std deviation

Then use the NDistribution table to get percentile, then is that the wrong thing to do? Or does percentile mean something else here? Since this uses mean, not median, as the 50th percentile, where x = mu and Z-score is 0.
Reply With Quote
  #2  
Old 02-05-2011, 12:54 PM
UFActuary UFActuary is offline
Member
 
Join Date: Jul 2005
Posts: 2,015
Default

Or is there another distribution table for skewed distributions? Isn't this because I am using a normal distribution for a sample that is not normally distributed, and that's the problem?

This is embarassing, but what I'm referring to are Superbowl Square bets. Where you bet on say Packers to have score ending in 7, Steelers to end in 2. And since there are 100 squares, the mean percent chance of winning is 1% assuming one payout at the end of the game.

BUT the median is much closer to .39%, this because the squares like 7,0 and 0,7 are biggies...they are as much as 10 times the percent likelihood to win as other squares.....

so if I take the percentage of my 7,2 which comes out to lets say 0.51%, I am above the median but well below the mean.

If I go 0.0051 - 0.01 / Std Dev I will get a percentile score below the 50% because I'm below the mean... is this logical? Or how SHOULD the percentile score be calculated mathematically?

Last edited by UFActuary; 02-05-2011 at 12:59 PM..
Reply With Quote
  #3  
Old 02-05-2011, 01:04 PM
Monkey D. Luffy's Avatar
Monkey D. Luffy Monkey D. Luffy is offline
Member
 
Join Date: Aug 2006
Location: 魚人島
Studying for HOLY CRAP DID U SEE THAT?
College: Not Zoidberg
Favorite beer: Anything. Now. Thanks!
Posts: 626
Default

Order all of the boxes in order of least likely to hit to most likely to hit, and then wherever your box comes in that order is the percentile it's in.

50th percentile is median.
Reply With Quote
  #4  
Old 02-05-2011, 02:08 PM
abt5 abt5 is offline
Member
CAS
 
Join Date: Dec 2010
Favorite beer: ice ice baby
Posts: 5,458
Default

Quote:
Originally Posted by UFActuary View Post
What I don't understand is if the median and mean are not the same (there's a skew).....in a sample

Then the amount by which your score is higher than half the sample, would be the median, NOT the mean.

So then if you calculate percentile by taking x - mu/ std deviation

Then use the NDistribution table to get percentile, then is that the wrong thing to do? Or does percentile mean something else here? Since this uses mean, not median, as the 50th percentile, where x = mu and Z-score is 0.
" x - mu/ std deviation" transforms everything into a standard normal. The mu here has nothing to do with the percentile itself, it's just used to give you the z-score. Once you have the z-score, you get the percentile from the standard table.
50% is always the median
Reply With Quote
  #5  
Old 02-05-2011, 02:21 PM
Gandalf's Avatar
Gandalf Gandalf is offline
Site Supporter
Site Supporter
SOA
 
Join Date: Nov 2001
Location: Middle Earth
Posts: 29,098
Default

From the distribution, create a table of the distribution function.

E.g. F(a) = 0
F(b) = 0.01
F(c) = 0.25
F(d) = 0.40
F(e) = 0.50
F(f) = 0.62
F(g) = 0.87
...
F(z) = 1.00

You may have a lot more values whose distribution function you calculate. For example if we took a group of adult males and ordered by weight in pounds, we might have values for F(120), F(121), F(122)...,F(250), and if our scale was sensitive we might have F(160.138) even.

Going back to my original example, since F(e)=0.50, then e is the median.

If I wanted to know what percentile f is, the answer would be .62 since F(f) = .62.

If I wanted to know what is the 40th percentile, I would say d, since F(d) = .40.

All that is true for any distribution, skewed or not. The mean could equal e, or be much bigger than e, or much less than e. Doesn't matter. All the percentiles are based just on the distribution function.

If you don't know enough about the distribution to know what the cumulative distribution function is, you might have to estimate it based on whatever you do know about the distribution. The normal approximation might be something you would use in those estimates. The estimates might or might not be any good.
Reply With Quote
  #6  
Old 02-05-2011, 03:15 PM
PhildeTruth PhildeTruth is offline
Member
CAS
 
Join Date: Dec 2009
Favorite beer: Allagash White
Posts: 3,236
Default

lol?
Reply With Quote
  #7  
Old 02-05-2011, 03:26 PM
r. mutt's Avatar
r. mutt r. mutt is offline
Member
 
Join Date: May 2002
Posts: 8,601
Default

Quote:
Originally Posted by PhildeTruth View Post
lol?
no question mark needed imo
__________________
(\ (\
( ^_^)
(_(")(") It all comes full akchigs eventually. ~ Full-On Devi
Reply With Quote
  #8  
Old 02-05-2011, 03:30 PM
Jasper07734's Avatar
Jasper07734 Jasper07734 is offline
Member
 
Join Date: Jun 2008
Posts: 5,700
Question

What's the name for the 42nd percentile?
Reply With Quote
  #9  
Old 02-05-2011, 03:35 PM
whisper's Avatar
whisper whisper is offline
Member
CAS AAA
 
Join Date: Jan 2002
Location: Chicago
Favorite beer: Hefewizen
Posts: 29,508
Default

VaR

----

The normal distribution has some nice properties which create easy ways to think of the distribution. The thing to remember is that these properties are unique to the normal distribution. Trying to take extrapolate the normal distribution properties to non-normal distributions is just going to cause problems - if not be flat out wrong.
Reply With Quote
  #10  
Old 02-05-2011, 05:25 PM
r. mutt's Avatar
r. mutt r. mutt is offline
Member
 
Join Date: May 2002
Posts: 8,601
Default

Quote:
Originally Posted by whisper View Post
VaR
Oof. I laughed at an actuarial joke.
__________________
(\ (\
( ^_^)
(_(")(") It all comes full akchigs eventually. ~ Full-On Devi
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


All times are GMT -4. The time now is 09:48 AM.


Powered by vBulletin®
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
*PLEASE NOTE: Posts are not checked for accuracy, and do not
represent the views of the Actuarial Outpost or its sponsors.
Page generated in 0.25721 seconds with 7 queries