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
- /
- Test for significant difference in proportion of M...

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

05-17-2012 01:19 PM

I am trying to determine if the proportion of Males over the time period of 2005 to 2010 changes significantly. Below is a breakdown of the proportions of gender over the years time period. I am having issues determining the best process to use to conclude if there is a significant change over time in the proportion of Males for my study. Thank you for any help you can provide.

I FIRST CREATED A DATASET FROM THE PROC FREQ OUTPUT

proc freq data=demographics;

tables gender/out=gender noprint;

by year;

run;

PERFORM A TTEST TO DETERMINE IF YEAR PROPORTIONS OF MALES DIFFER - IS THIS CORRECT?

proc ttest data=gender;

var percent;

where gender = "M";

run;

PERFORM A GLM MODELING YEAR AS MY OUTCOME - IS THIS CORRECT?

proc glm data=gender;

model year=percent;

where gender = "M";

run;

year | Gender | Frequency | Percent of Total |

Count | Frequency | ||

2005 | F | 17248 | 48.6902 |

2005 | M | 18176 | 51.3098 |

2006 | F | 16791 | 49.2968 |

2006 | M | 17270 | 50.7032 |

2007 | F | 17027 | 49.6356 |

2007 | M | 17277 | 50.3644 |

2008 | F | 17480 | 49.3409 |

2008 | M | 17947 | 50.6591 |

2009 | F | 19291 | 49.2733 |

2009 | M | 19860 | 50.7267 |

2010 | F | 22624 | 49.2619 |

2010 | M | 23302 | 50.7381 |

Accepted Solutions

Solution

05-17-2012
02:08 PM

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

05-17-2012 02:08 PM

Neither of the approaches that you showed are correct because they don't take the sample size into account. There are several ways to approach this.

- you could simply do a trend test in the PROC FREQ. Add the TREND option to the TABLES statement. This assumes that year is ordinal scale.
- you could do a logistic regression with gender as the outcome and year as the predictor. This assumes that year has a ratio scale.

Doc Muhlbaier

Duke

All Replies

Solution

05-17-2012
02:08 PM

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

05-17-2012 02:08 PM

Neither of the approaches that you showed are correct because they don't take the sample size into account. There are several ways to approach this.

- you could simply do a trend test in the PROC FREQ. Add the TREND option to the TABLES statement. This assumes that year is ordinal scale.
- you could do a logistic regression with gender as the outcome and year as the predictor. This assumes that year has a ratio scale.

Doc Muhlbaier

Duke

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

05-17-2012 02:36 PM

Thank you for your suggestions. I do realize now that I was not taking into account sample sizes.

I ran the suggested analysis and they do tell me if the proportions of gender change over time. Thank you again.

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

05-18-2012 12:26 PM

You are welcome. One caution is in interpretation. You have a very large sample size, so the test may be statistically significant without being "important". You observed a 0.6% drop between 2005 and 2006 and no more than a 0.25% after that; which may or may not be "important" depending on the context.