On Mon, 03 Nov 2003 11:10:20 +0000, John E wrote: > "R. Martin" <[EMAIL PROTECTED]> wrote in message > news:[EMAIL PROTECTED] >> John E wrote: >> > >> > Where can I information specifically on trends and patterns deduced >> > from histograms? I have to write a report based on this form of data >> > and I'm unsure of exactly what to do. >> >> To get an answer I think you need to define what you mean by "trends >> and patterns". Histograms collapse all the temporal information and so >> trends can not be discerned from a histogram. A series of histograms >> plotted from data obtained at different times will show trends in how >> the histogram changes, if that is what you are referring to. > > Yes I think it is. Here is a sample of the data I have to use in CSV > format. Is there anything that stands out in this data that you think I > should comment on - particularly in relation to a histogram?
<Data Snipped> It would help develop some intuition to better understand what your data is measuring. Is this across three hospitals or three groups at the same hospital? Is there any relationship between the groups/measures (ie. are they ordered in some fashion given the equal number of measures or are they totally independent)? Is there is any 'a priori' hypothesis or is this simply an exercise to use histograms to visually determine "high-level" differences between groups? Approaching this in a totally post-hoc (risky) fashion: I posted a quick histogram using density (not counts) for the data at: http://www.medanalytics.com/hist.png and a counts based histogram at: http://www.medanalytics.com/hist2.png For both of the above, the x and y axes are scaled evenly and the breaks are spaced the same for each set of 3. Another (better) review of the data visually would be a grouped boxplot, which I quickly generated and is at: http://www.medanalytics.com/boxplot.png The boxplot better demonstrates differences in the groups, including the high outliers in Group C. >From a more empiric, but still post hoc perspective: Pairwise t tests (with and without correction/pooled variances) show that A/B and B/C are different, but not A/C. A Kruskal-Wallis non-parametric test also supports differences across the groups. Paired Mann-Whitney tests did show differences across each of the three pairings. I also did a normal Q-Q plot of each for further review, the results of which should not be surprising, given the prior plots: http://www.medanalytics.com/qq.png There appears to be "something" fundamentally different about group/hospital C, but without more information (ie. are the three groups sampled differently), it is difficult to understand what that might be. I'll leave the plots up for a few days. HTH, Marc Schwartz . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
