turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- How to test the fit of discrete distributions?

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-09-2014 02:43 PM

I appreciate if someone can help me with this question.

Context:

In-field insect density was determined for different crop fields ("groups"). I have the frequency of the number of insects per sampling unit.

For instance,

field=1: the frequency of "zeros" insects per sampling unit was 353. The frequency of "ones" was 19.

field=2: the frequency of "zeros" was 580. The frequency of "ones" was 30. The frequency of "twos" was 3.

field | class.insect | observ_freq |

1 | 0 | 353 |

1 | 1 | 19 |

2 | 0 | 580 |

2 | 1 | 30 |

2 | 2 | 3 |

I would like to compare the observed frequency of each one of my fifteen fields with discrete distributions using chi-square test.

I am trying to use PROC GENMOD because I think PROC DISCRETE is not available anymore.

**proc genmod; **

**by field;**

**freq observ_freq; **

**model class.insect = / dist=binomial; **run;

I am planning to test for the following discrete distributions:

Binomial

Inverse Gaussian

Multinomial

Negative binomial

Normal

Poisson

Zero-inflated Poisson

Zero-inflated negative binomial

Double Poisson

Neyman type A

Final table should look like this image from Env. Ento. 30(2) :

Thank you.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-10-2014 08:08 AM

The chi-square value is presented in the goodness-of-fit table, so getting this into a dataset with ODS OUTPUT should not be difficult. I do worry that the goodness of fit applies to a model with some terms in it, so you may have to change to:

**proc genmod; **

**class field;**

**freq observ_freq; **

**model class.insect =field / dist=binomial; **run;

And of course that makes it impossible to find out if different distributions are better applied in different fields. So, try your method first, and if it gives no results in the goodness of fit table, you could try regressing on a dummy variable, such as regressing on a constant.

Steve Denham