Importing Histogram bin width & bin frequency data without access to underlying raw data

I have bin min/max and bin frequency data from a histogram, but I do not have access to the original data from which the histograms were prepared. What is the best way to format the data in a spreadsheet and then import it to MATLAB in a way that I can use the Distribution Fitter App? Currently I have the data in a series with column 1 being the bin max values and column 2 being the number of counts in that bin. For example below, a series of bins "x" 45 units in width, the first bin 510-555, the second bin 555-600, etc with the number of obsevations in the bin given in the second column "count":

    X  Count 
  555  15 
  600  22 
  645  49 
  690  81 
  735  104 
  780  133 
  825  162 
  870  182 
  915  192 
  960  158 
 1005  139 
 1050  97 
 1095  65 
 1140  52 
 1185  31 
 1230  10 
 1275  10 
 1320  3 
 1365  4 
 1410  3 

Answer 1

[dpb]

Just read the data in as is; no need for Excel. I'd use bin midpoints altho will make little difference on fitting is probably more intuitive in use. If the bin width is fixed at 45, subtract 22.5 from the max values.

[JOHN MOTZI]

Thanks. It will of course change the outcome a little as you say. Is there a way to enter it as a cumulative distribution function? It would be easy for me to reconfigure the data that way and it would more accurately portray the data.

[dpb]

You can compute whatever you wish from the data you have once you have it inside ML. What's the fixation on doing something first if this is the only data you do have access to? "Just start with what you got..."

[JOHN MOTZI]

My only fixation is trying to understand what can and cannot be done in ML an dthen how do do it - I am a new user so I really don't know what is possible.

[dpb]

MATLAB is a general purpose programming language -- there are essentially no restrictions to what you can do numerically in the language itself and you can always call mex files in C/C++/Fortran to do whatever in regards to systems calls, etc., etc., that aren't possible in ML syntax itself.

IOW, if you can think of what it is you want to compute, in all likelihood ML will let you compute it...or at least to a proof of concept level.

Of course, ML can't compute something from what it doesn't have access to...if you need an estimator that can only be computed from the raw data (such as a recent Q? the person wanted to compare distributions via t-test having already reduced the data to means--that's just simply not a computable statistic without the observations; Matlab has nothing to do with it).

You've really not told us anything here to go on as any specific wish...other than mention distribution fitting. Some pieces of ML are based on Toolboxes outside the base product so how much appears to be builtin is dependent upon just what toolboxes you have.

[JOHN MOTZI]

Thanks DPB! Fortran - now that's something I haven't done in years :)