A Review of SPSS 12.01, Part 2

Article excerpt

SPSS for Windows. North American price: $1,145 commercial, $599 academic. Extension modules start at $499. A variety of pricing programs are available for server licenses, multiple-unit sites, network, and maintenance agreements. SPSS, modules, and information can be obtained at www.spss.com/store or by calling 800-543-2185.

Note: At the time Part 1 of this review of SPSS software was written, SigmaStat and SigmaPlot were both products of SPSS, Inc. That is no longer the case; they are both now Systat products. As a consequence, I have limited this Part 2 review to SPSS version 12.01 alone.

SPSS has been on the market for several decades. It has been recognized as a premiere statistical software package during this entire time. In fact, I was part of a team of statisticians who engaged in a general review of statistical software for PC Magazine in 1993. Team members separately evaluated one to three packages, wrote of their findings, and collectively decided on the "best" overall statistical package. Nearly every commercial package was included in the testing. Criteria included ease of use, technical support, ease of installation, scope of capabilities, user interface, graphical components, and so forth. SPSS earned the magazine's top award. If the same statistical software evaluation test were to be held today--in 2004--I suspect that SPSS version 12.01 would be in close running for top spot again. But this does not mean that SPSS has no limitations or drawbacks. I shall discuss some of these in this part of the review.

In Part 1 (Hilbe 2003) I detailed the various capabilities of the SPSS Base module, as well as those of each of the extension modules. I also pointed out that most researchers find the Base Module sufficient for all of his or her requirements. It includes a truly extensive variety of descriptive statistics and excellent graphical capabilities, including 3-D rotations, ROC diagnostics, a wide choice of probability plots, as well as the usual quality control, histogram, scatterplot, boxplots and so forth. In addition, the Base module provides the user with the ability to perform ANOVAs, linear regression, curve estimations, a number of nonparametric tests, and factor and cluster analysis. SPSS data and file management facilities are, as would likely be expected, part of the base module. It is even possible to write one's own statistical routines or procedures in SPSS's programming language. However, I do have a criticism regarding a basic statistical routine that is not part of the Base module.

Statisticians who require logistic regression in their work need to acquire the Regression Models module at an extra cost. At the present time, I would think that logistic regression, as a commonly used routine across many disciplines, would be considered a Base model, and not one for inclusion in a specialized module.

SPSS is a modularized software package. Extension modules have a wide range of offerings; I am hard-pressed to think of other packages with greater overall scope. SAS may be the exception. However, with all of this capability, it lacks certain basic statistical routines.

For example, Poisson regression is not recognized as a capability in any of the extension modules. However, it can be accomplished in SPSS by the sophisticated user. Technical support, which I found to be extremely helpful, provided me with the directions and the logic underlying how the GENLOG routine can be "tricked" into doing a Poisson regression, parameterized in terms of either counts or rates. However, because GENLOG treats each case as a separate cell, continuous variables must be treated differently than categorical variables. Models having both continuous and categorical predictors result in special model construction concerns. It would be much easier for the user to have a routine devoted specifically to Poisson regression, or to have a generalized linear models (GLIM) routine with the traditional Poisson family option. …