How do you know which distribution to use for your problem?

Click For Summary

Discussion Overview

The discussion revolves around the selection of appropriate statistical distributions for various problems, focusing on the criteria and methods for determining which distribution to use among several options, including Binomial, Negative Binomial, Hypergeometric, Poisson Distribution, and Poisson Process. The scope includes theoretical considerations and practical applications in data analysis.

Discussion Character

  • Exploratory, Technical explanation, Debate/contested

Main Points Raised

  • One participant suggests that the choice of distribution may depend on fitting data to a model, where the distribution yielding the least error becomes the working model.
  • Another participant proposes that one might guess the distribution based on first principles, particularly in physical problems.
  • A different viewpoint mentions the possibility of tracing distributions through established analytical relations, noting that certain combinations of distributions lead to others, such as the sum of normally distributed variables being normally distributed.
  • It is also noted that certain types of experiments are typically associated with specific distributions, such as the binomial distribution for coin tosses and the Poisson distribution for customer arrivals within a time frame.

Areas of Agreement / Disagreement

Participants express differing views on how to determine the appropriate distribution, with no consensus reached on a singular method or approach. Multiple competing perspectives remain regarding the criteria for selection.

Contextual Notes

Some limitations include the dependence on the context of the problem, the potential for approximation under certain conditions, and the need for further clarification on the assumptions behind each proposed method.

sjaguar13
Messages
49
Reaction score
0
If you have a problem that involves some distribution, how do you know which one to use? The ones we covered so far are:
Binomial
Negative Binomial
Hypergeometric
Poisson Distribution
Poisson Process
 
Physics news on Phys.org
You don't. You either take data and fit the distribution to one of those (and whichever one gives you the least error becomes the working model), or you guess the distribution from say first principles (say if you are a physicist and working on a physical problem).
 
Sometimes you may be able to trace distributions as "precedents" or "dependents" by uisng established analytical relations between distributions. E.g., if two variables are normally distributed, then their sum is also normal, and their quotient is Cauchy. The logarithm of a Normal variable has a Log Normal distribution. And, sometimes you can approximate (distributions approach one another under certain limit conditions, so it may be possible to use a simpler distribution in place of a more complex one).
 
And at some level, you do know. For example, you know whether it's a discrete or a continuous distribution. Also, we know that centain types of "experiments" are associated with certain distributions. Thus, a coin toss is associated with binomial; "consumer choice" is associated with multinomial; the number of balls in each of N boxes is associated with Chi-square; the number of customers arriving "between 9AM and 10AM" is associated wtih Poisson.
 

Similar threads

  • · Replies 6 ·
Replies
6
Views
2K
  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 12 ·
Replies
12
Views
5K
  • · Replies 3 ·
Replies
3
Views
3K
  • · Replies 7 ·
Replies
7
Views
2K
  • · Replies 4 ·
Replies
4
Views
2K
  • · Replies 15 ·
Replies
15
Views
2K
  • · Replies 9 ·
Replies
9
Views
3K
  • · Replies 7 ·
Replies
7
Views
2K
  • · Replies 16 ·
Replies
16
Views
2K