Re: Histogram "Bins" tab options problem

[Andreas Guelzow <aguelzow pyrshep ca>]

On Tue, 2011-02-08 at 21:54 -0500, Daniel P. Dougherty wrote:
> Ok, in my version (1.10.12) the number formating only works for the Histogram 
> Chart type. 

> If you select "Column chart" option you still get the "above ..." 
> and "up to ..." stuff but the X axis in the column chart only treats them as 
> words (text).

I see what you are trying to do. You need to do the formatting on the
results on the sheet and the create the column chart from there. The
column chart just uses the formatting from the sheet.

Of course the column chart should allow you to change teh formatting
afterwards. But that has nothing to do with those cells containing
"text". They do not!

> Consider the following (discrete) example data:  
>      2
>     10
>     10
>      5
>      9
>      2
>      5
>     10
>      8
>     10
>
> For interest of comparison with another popular software I used MATLAB's 
> hist() function to produce the first graphic (mlabhist.png).  This is a 
> frequency histogram so the Y-axis shows bin count and it is apparent that  no 
> normalization of the bin heights was done.
>
> Next using Gnumeric (1.10.12) I go to Statistics->Descriptive Statistics-
> > Frequency Tables->Histogram and select 10 bins (default bin style) and under 
> Graphs & Options I select "Histogram chart".  I get the second graphic 
> (gnmrchist1.png).  Here the heights have been (silently) normalized.

There was noting silent about that. You purposefully did not chose a
column chart but a histogram chart, ie. a column chart with normalized
frequencies.

>    For 
> comparison I did same but with "Column chart" option (see last image 
> gmrchist2.png) with 0% Gap between bins.  Now the Y-axis values are same as 
> MATLAB but the x-axis label is mal-formed text.  Here the heights were not 
> (silently) normalized.  So it's a different analysis result, behavior and look.
>
> In the end, I found it difficult (to me) to reproduce the default MATLAB style 
> histogram and also to get nicely formated x-axis labels in Gnumeric.

This would be really simple:

1)  run the analysis
2)  format the spreadsheet cells however you like
3)  then create your column chart.  

>   The 
> conversion of the text labels

There are no text labels. The spreadsheet cells contain numbers
(formulae that result in numbers). 

Andreas

>  to numbers as you had mentioned does not occur 
> in the Column chart type and the use of scientific notation as the default 
> number formatting is (apparently) not consistent across the different "Graphs & 
> Options" chart types. 
>
> Yes now that you have explained things, I can understand why things behave the  
> way they do, but I think most users (and students) will get thrown by some of 
> this.  My sense is that the learning curve to get in and do a quick basic 
> (perhaps a student's first) frequency or relative frequency histogram in 
> Gnumeric has just gotten a bit _higher_ in this new version and users will get 
> frustrated with that.  


> On the other side the more advanced users of Gnumeric might be pleased to see 
> an "Empirical PDF/CDF" option eventually appear under  Statistics->Descriptive 
> Statistics->Frequency Tables where the (normalized) Histogram Chart object can 
> be overlayed by a smoothed kernel density estimate plot.  That would be 
> similar to functionality one can only get now by combining MATLAB's ksdensity 
> and ecdfhist functions (see last image).  That would be cool!