Free statistical software

Free statistical software is a practical alternative to commercial packages. In general, free statistical software gives results which are the same as the results from commercial programs, and many of the packages are fairly easy to learn, using menu systems, although a few are command-driven. These packages come from a variety of sources, including governments, non-governmental organizations (NGOs) like UNESCO, and universities, and are also developed by individuals.

Some packages are developed for specific purposes (e.g., time series analysis, factor analysis, calculators for probability distributions, etc.), while others are general packages, with a variety of statistical procedures. This article is a review of the general statistical packages.

Sources of free statistical software
Some of the free software packages are from governmental or NGO organizations, such as Epi Info, from CDC, and IDAMS from UNESCO. Some other software packages are from smaller or independent organizations or universities, such as Instat or Irristat. Another package, the R project is being developed by a group of volunteer individuals. A large proportion of free statistical software packages, however, are from individuals. Some of these software packages from individuals include Easyreg, MicrOsiris , OpenStat , PSPP , and Zelig.

In some cases, the statistical software packages were developed for the purposes of making key technologies available to those who could not otherwise afford them, to empower development,. In other cases, the packages were developed as teaching aids,. Other packages were developed for specific purposes but can be more generally used. Examples are Irristat, developed for agricultural analysis, and Epi Info , developed for public health. A couple of packages don't appear to give any statements about why they were developed, other than just general use for statistical analysis, ,.

Reviews of free statistical software
There are a few reviews of free statistical software. There were two reviews in journals (but not peer reviewed), one by Zhu and Kuljaca and another article by Grant that included mainly a brief review of R. Zhu and Kuljaca outlined some useful characteristics of software, such as ease of use, having a number of statistical procedures and ability to develop new procedures. They reviewed several programs and identified which ones, at that time, had the most functionality. At that time, several of the programs may not have had all of the desired ability for advanced statistics. Grant reviewed some of the programing features of R, and briefly mentioned the availability of other programs. One other paper reviewed statistical packages, mainly commercial, but includes R. One article reviewed EasyReg and included a discussion of it's accuracy.

Only one review has compared the output of various packages. In this review, all of the packages read either CSV (Comma Separated Values - text files in which all values are separated by commas) files or excel format. All of the packages gave exactly the same results for correlation and regression. The free software packages also gave the same regression results as did excel. One of the main differences among the packages was how they handled missing data. With the example data sets used in the review, and for the package versions available in November 2006 when this review was conducted, two packages, MicrOsiris and Epi Info, could read files with blanks for missing. Two other programs, Stat4U and WinIdams need something for the missing, like -9 or -9.99. The other packages could only handle data sets with no missing values.

Two websites that list software also have very brief reviews of each package. These two sites are StatCon and by Pezzullo. These sites mainly offer a brief list of the features available in the packages. Similarly, one bachelor's thesis compares the statistical procedures available on free statistical packages. In this review, R had all of the procedures, OpenStat had 16, MacAnova had 15, and Microsiris had 12. The others had from 8 to 11 of the procedures.

There is also a journal specifically for statistical software, although the main focus is on commercial software, R and some coding snippets.

These free software packages have been used in a number of scholarly publications, so that at least various journals, NGOs or other organizations regard the packages as valid. For example, OpenStat was used in a research letter to JAMA and in several published studies, ,. Irristat is used in this agricultural report, EasyReg is listed or used in these papers , , , various versions of EpiInfo were used in these papers , , and WinIdams was used in these papers ,.

While Microsiris doesn't appear to be used in academic research, the author of the program was one of the original authors of OSIRIS, which was the starting program from which WinIdams was developed. The author of Microsiris also has also contributed or co-contributed several components to WinIdams.

Using free statistical software
Before using any statistical packages, it is generally a good idea to have a solid background in Statistics. Then the packages can be used to the best advantage, for example, to choose the most appropriate test, to make sure all the necessary assumptions are met, so that the appropriate conclusions can be drawn.

Once the statistical issues are understood, the next step is to decide which package to use. Most of these packages are menu driven, and can be learned a couple of hours at most, except R, which is generally code driven and requires a much longer time to learn, and to some extent CDC's Epi Info, which also takes some time to learn.

Several of the packages also have tutorials. These tutorials help with a basic introduction and learning the basics of programs. For example, CDC has these tutorials about Epi Info,. The CDC page also lists a video slide show tutorial from the University of Nebraska, and another site has on line training classes. R has a large number of tutorials and manuals, in English and other languages, , , and a faq site. A few of the packages have a email discussion lists including R, OpenStat , and PSPP.

Most of the packages have on line manuals, guides or help pages. These manuals or guides are useful when there are questions about specific procedures or statistical tests. Some manuals or guides are for R, EasyReg , OpenStat , PSPP , Vista , WinIdams , , Microsiris and Zelig. The CDC EpiInfo site itself does not have a manual, but one faculty member from Emory's School of Public Health has an introductory manual.

Menu driven packages
Many of the packages have some kind of opening menu that is used to get or enter the data, manipulate the data, and select the statistical analysis. One example, from Microsiris, is this:

In general, users would import data, from text, a spreadsheet or from some other form. The menus then allow the user to do things like manipulation of data, like new calculations, selecting certain cases, creating new variables, and so on. Finally, users can click on the statistical procedure and get output.

Command driven packages
A few programs, like WinIDAMS and R need commands for many of their procedures.

Getting data
Most packages are able to import data from excel or CSV (text with commas separating values).

One consideration is whether there are missing data. Some packages, like PSPP and MicrOsiris, can automatically deal with the missing data. So for example, say one set of data look like this:

In this data set, Sam is missing age, and Sally is missing whether she was born in the USA. When some packages, like PSPP or MicrOsiris, read in or import the original data set, the packages will recognize that those values are missing, and do their calculations accordingly.

Other packages need a 'place holder', such as '-9' where there is missing data. Before the package is used to read the data, the data set has to be edited to put in '-9' where there are missing data. So for example:

The data set includes '-9' and then people who are reading in the data need to tell the program that the -9 means missing data.

Limitations of packages
Most of the packages have limitations of some sort. Variables in WidIDAMS are limited to 9 digits in length and so have to be manipulated before analysis. In the version of PSPP current as of April 2009, there are a limited number of procedures available, including means, frequencies, crosstabs, two non-parametric tests, t-tests, anova and basic regression. In addition, the output is, apparently, not easy to use as it cannot be copied and pasted to other applications, and it is not clear where, in Windows Vista, the output is saved. Several of the programs, including Easyreg, Epidata and Instat, do not appear to handle missing data or do not handle it well. While EpiInfo has many statistical procedures, correlation is not one of them. Rather correlation is found by regression. This means that EpiInfo will not produce a single table showing correlations among multiple variables. According to the Zelig installation manual, use of Zelig requires that R and several of it's libraries already be installed. Installation also requires some degree of background in R.