{"id":1259,"date":"2016-03-09T20:48:33","date_gmt":"2016-03-09T19:48:33","guid":{"rendered":"https:\/\/blogs.bmj.com\/adc\/?p=1259"},"modified":"2016-02-24T16:58:39","modified_gmt":"2016-02-24T15:58:39","slug":"statsminiblog-middling-differences","status":"publish","type":"post","link":"https:\/\/blogs.bmj.com\/adc\/2016\/03\/09\/statsminiblog-middling-differences\/","title":{"rendered":"StatsMiniBlog: Middling differences"},"content":{"rendered":"

\"calculator-97842_640\"Is there a difference between these two non-parametric groups<\/a>? (For example, the duration of cough after bronchiolitis…)<\/p>\n

 <\/p>\n\n\n\n
 <\/p>\n

Time to cough resolution (days)<\/p>\n

 <\/td>\n

\u00a0Rx A:\u00a0Median 15 days<\/td>\n\u00a0Rx B: Median 15 days<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

You’d want to see a stats test though<\/a>, wouldn’t you. So what if it was ..<\/p>\n\n\n\n
Time to cough resolution (days)<\/td>\n\u00a0Rx A:<\/p>\n

Median 15 days<\/td>\n

\u00a0Rx B:<\/p>\n

Median 15 days<\/td>\n

\u00a0Median Difference<\/p>\n

1.0 days (95% CI -1.0 to 2.0)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Hang on! You might say. Those medians are identical – how can that median difference be 1.0 ?<\/p>\n

 <\/p>\n

Well – there’s a difference between a\u00a0Median Difference and a the difference between the group medians. The latter is what it sounds like, and is tested with them non-parametric testy <\/a>things. The former is trickier …<\/p>\n

Take the durations of cough and put them side by side.<\/p>\n

Now take the difference between those pairings you’ve got.<\/p>\n

Now rank the resulting\u00a0“difference between those pairings”, and take the 50%ile.<\/p>\n

That’s the median difference (the median of the differences).<\/p>\n

Statisticians, huh?<\/p>\n