SPSS more than a pretty face

Version 7.0 of SPSS Inc.'s eponymous package puts statistical power in a gray flannel suit, meeting corporate expectations for polished presentations. However, technical users will still find SPSS' analytic tools satisfying.

Released just before year's end, the $695 SPSS 7.0's high-capacity design took roughly twice as long as Microsoft Corp.'s Excel (which is lean and mean at presenting statistical output) on tasks involving the small data sets that a memory-based spreadsheet can handle.

Excel outdoes SPSS in comfort as well as speed when it comes to automated aids for such tasks as importing ASCII data, making it easier for the Excel user to do things the fast way. SPSS users, though, will be tempted to keep life simple but slow.

Users who want more active guidance in choosing analytic techniques should look at products like Business Forecast Systems Inc.'s Forecast Pro. This package reviews the data set and recommends an appropriate time-series technique based on a library of both formal and informal rules.

But we know of no product that combines such active guidance with the analytic breadth of SPSS, which goes far beyond the specific domain of time-series forecasting to include every important family of classical statistical analysis.

Slow and steady

For our tests, we installed SPSS 7.0 on a 90MHz Pentium-based Micron Electronics Inc. P90 Peripheral Component Interconnect machine with 16M bytes of RAM, running Windows 95. Our test data records contained two randomly generated independent values and a perturbed linear combination of those values to simulate moderately correlated observations.

Excel opened our test file of 10,000 three-variable records about twice as fast as SPSS, with both programs using efficient fixed-column-width parsing of the input data. This actually understates Excel's practical advantage, because Excel automatically recognized the input file's format and offered us appropriate default values for data field boundaries. With SPSS, we had to manually enter start and end columns for each field, without even being able to view the raw data file to help us make those settings.

When we allowed SPSS to identify field delimiters (in this case, spaces) on the fly, SPSS took almost three times as long to process the input file as it did in fixed-width mode. This is clearly due to a parsing bottleneck, because SPSS read the same data in its native file format 30 times faster than from the ASCII file in fixed-width mode. Also, Version 7.0 was just as quick as Excel in opening a native-format file.

On the plus side, SPSS can handle far larger data sets than Excel, which is limited to 16,384 rows. SPSS imported a data set of 100,000 records without incident, though this took proportionately longer than reading the smaller data set mentioned above. For large collections of data, even those that need only simple analysis, SPSS can therefore make the difference between off-the-shelf GUI convenience and tedious custom programming.

Brains and beauty

The Output Navigator display in SPSS 7.0 captured the output from our test sessions, not in the typewriter-style transcript provided by earlier versions, but in an accumulating WYSIWYG-style formatted document that was indexed through a point-and-click hierarchical tree. The tree display let us quickly locate and display any portion of our transcript. In addition, by dragging and dropping, we could rearrange the sequence of items in our transcript or suppress details we weren't interested in at the moment.

This is a significant improvement from the limited hypertext integration of text output and charts that whetted our appetite in Version 6.1. We were entirely satisfied with SPSS 7.0's output management.

Version 7.0 also supports OLE (Object Linking and Embedding) 2.0. We could embed output by dragging and dropping it into documents created with OLE-compatible applications.

SPSS' output improvements aren't just cosmetic. The new output model has some brains behind its pretty face as well as an accommodating manner, providing a superbly designed set of pivot-table tools that gave us immediate access to different groupings of multidimensional results.

When we displayed the optional on-screen "pivot trays," SPSS 7.0 showed regions labeled Layer Dimensions, Row Dimensions, and Column Dimensions. The program noted each variable in our table by an icon with a pop-up label. We could quickly and intuitively drag these icons from one tray to another, or drag them into a different sequence, and easily examine the same data in any of several multipage or nested single-page layouts.

Graphical data plots are well-integrated into the new Output Navigator. An animated skeleton of axes let us quickly adjust three-dimensional perspectives without the CPU burden of replotting our entire data set. We found this at least as effective a compromise as any other graphing tool that we've used.

SPSS 7.0 also retains such useful features as graphical identification of outlying data points.

Online tutorial aids and help screens in SPSS 7.0 have a radically new look, far superior to the sleep-inducing tutorial that we castigated in Version 6.1.

Menu titles, however, still describe a family of techniques rather than anticipating a user task. For example, the Time Series menu bears such choices as Autoregression, ARIMA, and X11ARIMA. However, the menu titles are offset by readily accessible help screens with clearly written explanations of these analytic techniques. Most users will easily figure out which techniques fit various situations.

SPSS, of Chicago, can be reached at (800) 543-2185 or at http://www.spss.com.

Copyright (c) 1996 Ziff-Davis Publishing Company. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of Ziff-Davis Publishing Company is prohibited. PC Week and the PC Week logo are trademarks of Ziff-Davis Publishing Company. PC Week Online and the PC Week Online logo are trademarks of Ziff-Davis Publishing Company.