Dear Leonhard Mada, I think I now understand what you intend. You really want a "closed" interface for naive users of statistics, probably through a menu driven interface similar to the analysis toolkit in Excel. The problem of course is what methods to offer. This kind of integration means that from the user's point of view you have written a complete statistics package. What is not on the menu is not available to the user. Using an established menu package like RCommander has the advantage of building on the work somebody else has done already.
Not let me add some more comments: > 1. (for comment c) > I encountered often variables that were NOT limited to a single column: > e.g. age was written both in columns A, B, C and D. Therefore, I would > like to be able to select a range individually for every variable: > age: A1:D50 > Similarly, Blood Pressure: H1:I100 > > The A1:H100 selection won't work in this case, because this is not a > data frame as the same data is split into columns. Rewriting the data > into a single column is time-consuming (and the data is often written in > more than one column because of another factor, so rewriting it into a > single column will loose some information). > I think supporting variables which are split into noncontagious areas is not something a good package should support. When data come in in a "mess" like you describe, one should push the user to struture them before doing the analyses. Having one variable distribute across multiple ranges raises the probability of errors quite a bit. Seeing the data in an organized way before doing analyses is something people doing statistics should get accustomed to. > 2. (comment b) > I had in mind the: > fisher.test( matrix( c(number_1, number_2, number_3, number_4), 2 )) > and similar contingency tables. Somebody who does NOT have any idea of > R, won't be able to perform such a simple test until he learns to > construct a matrix. BUT if the user only needs the fisher test, then , > you are right, there is no real need for the user to create a matrix on > his own; it is easy to select the contingency table and to pass the > correct command to R (without the user having to know more details about > matrices). In this case the question is if you want to create the contingency table in Excel and then have R only perform the additional computation. If you are using R for statistics, you also could send the raw data to R and have R compute the fisher.test. Of course, there might be situations where you only have the contingency table in Excel, but not the data. In that case, transferring a matrix (which should be in the transfer toolkit anyhow) and then applying fisher.test to the matrix does not really involve any deeper knowledge of the R language. > > 3. (comment a) > The previous comment applies to factors, too. I had primary in mind the > ANOVA test, but then again, the user does not need to know the details > of parsing the arguments and everything could be hidden in the > implementation. Somebody who needs factors for a different analysis, > will most likely know how to create them in R. This is roughly the same question as the first one. Do you want to build your own little statistics packages which uses the R computational engine? If you have any chance, take a look at the demos coming with RExcel. I think seeing examples of possible interfaces to spreadsheets in statistics might help avoiding duplication of work that has been done already. And here is my "favorite complaint" again. I have not tested OpenOffice 2.2 yet. DO you know if it graphics is finally fast enough to do slider controlled animation? -- Erich Neuwirth, University of Vienna Faculty of Computer Science Computer Supported Didactics Working Group Visit our SunSITE at http://sunsite.univie.ac.at Phone: +43-1-4277-39464 Fax: +43-1-4277-39459 --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
