Friday, April 23, 2010

04. Transparent My Computer USA, LLC

04. Transparent My Computer, originally uploaded by Criss.AC.

Microsoft Store

ArabicChinese (Simplified)Chinese (Traditional)DeutchEspanolFrenchItalianJapaneseKoreanPortugueseRussian

SYSTAT IS A FULL-FEATURED, dedicated statistical analysis program. Originally written for the first-generation IBM PC (I still use the DOS version 5, circa 1991, on my handheld computer), the program has grown into a sophisticated, easy-to-use product. The current version is Systat 11, released in 2004. Running only on the Windows (Windows 98 or later) platform, the hardware requirements are modest: any Pentium microprocessor, 64MB RAM and 150MB of storage (80MB less if you omit the PDF documentation). I used Systat on several computers, from a Pentium II to a Pentium IV. The speed performance on the Pentium II was, of course, slower, but still quick enough as to be unobjectionable.

Most users, when looking for statistics software, primarily want to know its capabilities (statistical and graphical), its accuracy, how easy is it to learn and how easy is it to use. Other issues would be the quality of its help system (written and on-disk documentation), vendor support and perhaps price. While a limited review cannot comprehensively answer all of the above issues, I will try to address many of the salient factors in each.


Systat's list of statistical features is broad. A partial list of Systat's statistical features is given in Table 1, and its graphical capabilities are summarized in Table 2.

Working with Systat

Systat installs like many Windows products do, asking for a serial number and giving the user a choice of what should be installed and where. Upon installation, like most users, I initially eschewed the product manuals and started playing with the product. The Systat screen has three regions: the Viewspace, Workspace and Commandspace. (I learned these later names after reading the first Systat book, "Getting Started.") Each of these spaces consists of specific windowpanes to do specific chores. Figure 1 shows a typical Systat screen with information in the spaces.

The most important of these spaces is the Viewspace. The Viewspace consists of three possible windows or views: the Data Editor (which shows a spreadsheet view of your data), the Output Pane (which displays the statistical or graphical output of your analysis) and the Graph Editor (which allows you to modify the look of your graph using Systat's Graph Editor, a program with much flexibility and power). The Workspace is a navigational path for your input requests and resulting output. The Commandspace is where you can type specific commands (ignoring the menu system) to get statistical output.

Getting data into Systat is easy because Systat imports data from many formats (including Excel, SAS, SPSS, Minitab, Statistica, Stata, S-Plus, JMP, ASCII and others). The program recognizes qualitative data, and these variables must have a name with a dollar sign at the end (e.g., Gender$). Systat automatically supplies this when it imports data. Unfortunately, it does not permit qualitative data to have more than 12 characters. Quantitative data is automatically displayed with three decimals (even if the source data was originally integer-valued), which the user can change to a maximum of nine, but unfortunately can only do so one variable at a time. I found that Systat sometimes, but not always, had trouble identifying data when using cut-paste. That is, I copied quantitative data from an Excel spreadsheet into Systat's spreadsheet, and Systat erroneously identified this as qualitative data.

In my statistics classes, I frequently require students to manipulate the data before running some analysis. Typically, they need some subset of the data, or they must make transformations. Systat has many data management features that let you do just about anything to the data prior to analysis. Besides the standard operations, such as List, Sort, Standardize, By Groups (which gives separate output for each value of some variable), Select Cases (a filter that analyzes part of your data based on Boolean conditions you specify) and many others, there is the Let and If-Then operations. With Let, you can use any combination of variables, numbers, conditions with Boolean operators, etc. to get the required value for an existing variable or a new variable. With If-Then you can transform variables conditionally.

It was while working with the data manipulation that I ran into difficulties using Systat and had to resort to conferring the manual. Systat originated as a command-driven program, and as newer versions evolved, many of the commands were incorporated into menu choices, but the "engine" that still drives the software is command language. Most, but not everything, is available by clicking on some menu choice. I found it easy to use the menu system to enter one condition in a data operation but not so for multiple conditions. The best recourse is to use commands.

The help system and the documentation give numerous examples, mostly emphasizing the commands needed. For example, using my data set about automobiles (culled from Consumer Reports), I wanted to analyze a specific subset of this data: American or Asian cars whose mileage per gallon was between 25 and 30 and whose reliability rating was either "Excellent" or "Good." I had trouble getting this subset of the data using Systat's menu system, but when I typed:

MILEAGE < 30) AND (RELIABLE$ = "good" OR RELIABLE$ = "excellent")in the Commandspace (bottom of the screen in Figure 1), I got the cars that meet these criteria, and subsequently used this filter to analyze them. I had similar problems with the If-Then and the Let transformations. In summary, for much of what I wanted to do, I found Systat's menu system easy and intuitive. However, for more elaborate transformations, the menu system is neither, and I was best served by typing commands.

What's wrong with command language? I see two possible drawbacks: a casual user may forget the specific syntax and thus have to look it up in Systat's Language Reference manual, and a user might not even know that a specific feature is available unless he/she finds it in the language manual.

Linear regression analysis is one of the mainstays of what I frequently do with my students. Getting the model and the output from Systat was easy; however, the output is relatively sparse and limited. Besides the standard regression output (model and related statistics, coefficients and their associated statistics, ANOVA table, correlation matrix of the coefficients, Durbin-Watson D statistic), a residual-predicted scatter plot and warnings of outliers, Systat gives the eigenvalues of the unit scaled XTX matrix, condition indices and variance proportions. The latter three items help the user detect the presence of collinear variables and determine which, and how much, model coefficients have been damaged.

There is no option for additional residual analysis per se; you must opt to save the residuals in a new file, retrieve that file and make whatever analysis you wish. Do you want 90 percent instead of the default 95 percent confidence intervals for the regression coefficients? Sorry. I was also disappointed to find that Systat does not find a prediction interval for a new observation. If you enter the new values for the independent variables in the spreadsheet, run the regression, save the residuals, you get a "sepred" or "standard error of prediction" variable value for this new observation. Knowing this number, and with the correct t-value, I knew that I could manually calculate the prediction interval limits for the new observation. Not quite. Unfortunately, the "sepred" value Systat gives is the standard error for the mean response and not the standard error for an individual response, something the documentation does not specify.

Systat is a pure "manual" product when it comes to time series. That is, you must specify a specific forecasting model (Brown's, Holt's, Winters', ARIMA, etc.); you must also give the smoothing constants, AR, MA values, etc. Systat does not find them for you. Getting the right model with the optimal parameters for your data can thus require a tedious trial-and-error process.

Systat's cross tabulation capabilities are extensive, offering a wide variety of statistical analyses. The only minor drawback is that Systat does not give a warning if conditions for some statistical test (e.g., Pearson's chi-square) are not met; it just performs the hypothesis test. Systat's multivariate features are also deep, and I encountered no difficulties or drawbacks.

Perhaps Systat's strongest feature-set is its graphics capabilities. Table 2's list of the main type of graphs just scratches the surface of the immense possibilities of what Systat can display. For example, Systat can make two- and three-dimensional plots. Figure 2 shows one such plot (I added the spikes to help show perspective). What does the multiple regression model (or smoother, as Systat calls it) for this data look like? Figure 2a shows the result.

If you want to show residuals, or label the data points, or plot by subgroups, Systat can do these and much more. Systat's CD has much sample data, including data to draw maps. The map chapter in the Graphics manual explains what data specifics (latitude, longitude, etc.) are needed to make your own maps and how to embellish them. The Graphics manual has 442 pages, a testament to Systat's graphics power and flexibility.

Systat's accuracy is excellent. I always got correct results for every file I threw at it, including many with problematical data.

The only significant "bug" I found was with data transformation. When sorting qualitative data in ascending (but not descending) order, my data "disappeared," and all I saw was a blank spreadsheet. Actually, Systat did not erase the data, but sorted and sent it to "Neverland" (specifically, row 499). Systat does offer periodic patches to the software on its Web site, but this anomaly persists.

Systat keeps a running log of the output of all your analyses in the Viewspace pane. This can be saved as a RTF file (easy to import into any word processor) or an HTML file.

Documentation and Help

SYSTAT'S DOCUMENTATION is excellent and follows the standard set by founder Leland Wilkinson two decades ago--it not only shows you how to use the software, but teaches the topic as well. For example, the opening sentences on page one of the graphics manual set the tone: "This chapter has nothing to do with Systat. It has everything to do designing good graphs." The next 20 pages give the user insight on the psychophysics of perception, visual illusions, graphic design, information overload and more. In all, there are seven Systat manuals (3,181 pages total), devoted to getting started, data, graphics, command language and three volumes on statistics. Written by several authors (many by Wilkinson, a superb expositor), they have been edited to give the seamless read of a single author. The context-sensitive online help is also excellent. Clicking on the help icon, Systat responds with an opening page explaining the options. There are also hyperlinks to "How to," "Syntax," "Examples," "More" and "See Also," each offering a list of appropriate possibilities to examine.


Summary of overall assessments of Systat:

Ease of Learning -- Very Good

Ease of Use -- Good

Breadth of Statistical Capabilities -- Very Good

Documentation and Help -- Excellent

Graphics -- Excellent

Vendor Web Support -- Excellent

Vendor "Live" Support -- Not tested

Systat is an accurate, powerful, nearly comprehensive statistical program with an excellent graphics accompaniment. Some features (quality control, multivariate statistics) are stronger than others (time series), and so the program may not be for everybody. I like it very much, and it is one my "top five" statistical programs that must be on my hard disk. I recommend you download the 30-day trial version and test it to see if it meets your requirements.

Jack Yurkiewicz is a professor of management science in the MBA program at the Lubin School of Business, Pace University, New York. Besides management science, he teaches business statistics and forecasting. His current interests include developing distance-learning courses for these topics and assessing their effectiveness.


Systat 11 is available from:

Systat Software, Inc.

501 Canal Blvd, Suite E

Point Richmond, CA 94804-2028


Phone: 800-797-7401

Fax: 800-797-7406


Web site:

The Web site lists additional headquarters for the United Kingdom, Europe and Asia.


Commercial: $1,299

Government: $999

Academic: $499

Upgrade prices from previous versions available.

Thirty-day full-trial version available on the Web site.


Basic Statistics
* Descriptive Statistics
* Confidence intervals for mean and proportion, hypothesis tests for
mean, proportion, standard deviation, correlation
* Resampling--Bootstrap, without replacement, Jackknife
* N- & P-Tiles: Cleveland, Weighted average 1, Weighted average 2,
Weighted average 3, Empirical CDF, Empirical CDF (average), Closest
Cross Tabulation and Measures of Association
* One-, two-way and multiway tables
* Row and column frequencies, percents, expected values and deviates
* List layouts, order categories, define intervals, including missing
* 2 X 2 tables: likelihood ratio chi square, Yates', Fisher's, odds
ratio, Yule's Q
* R X R tables: McNemar's test, Cohen's kappa
* R X C tables: unordered levels, phi, Cramer's V, contingency
coefficient, uncertainty coefficient, Goodman-Kruskal's lambda
* R X C ordered levels: rho, Goodman-Kruskal's gamma, Kendall's tau-b,
Stuart's tau-c, Somers' D
* Mantel-Haenszel test, Cochran test
* Linear (least squares, Stepwise, Bayesian, Ridge), Two Stage Least
Squares, Nonlinear, Robust, Logistic, Probit
* One-way ANOVA: multiple tests, Bonferroni, Tukey-Kramer HSD,
Scheffe, Fisher's LSD
* Two-way ANOVA: post hoc tests on least squares means (Bonferroni,
Tukey, LSD, Scheffe)
* Repeated measures: one-way, two or more factors, three or more
* Designs: unbalanced, randomized block, complete block, fractional
factorial, mixed model, nested, split plot, Latin square, crossover
and change over, Hotelling's T2
* Bootstrap, without replacement, Jackknife

General Linear Model
Time Series Analysis
* Moving Averages, Exponential Smoothing, Holt's Method, Winters'
* Box-Jenkins ARIMA model
* Fourier and inverse Fourier transforms

Nonparametric Statistics
Multivariate Statistics
* Cluster, Factor, Discriminant Analysis
Additional Statistical Capabilities
* Conjoint Analysis
* Survival Analysis
* Path Analysis (RAMONA)
* Perceptual Mapping
* Spatial Statistics
* Signal Detection Analysis


* Bar, Dot, Line, Pie, Profile and Pyramid Charts
* Histograms, Box, Dot and Density Plots
* Scatter, High-Low-Close, Probability, Quantile Plots
* Multivariate Graphs: Scatterplot Matrix, Parallel Coordinate Display,
Andrew's Fourier Plot, Icon Plot
* Function Plot
* Map Displays -- Geographic projections (2-D): Gnomonic, stereo,
Mercator, ortho, Lambert, Robinson, sinusoidal, Miller, Peters,
* Quantile and Probability Plots -- 28 probability densities

Source Citation
Yurkiewicz, Jack. "Systat 11: comprehensive statistical software and great graphics." OR/MS Today 32.1 (2005): 52+. Computer Database. Web. 23 Apr. 2010.
Document URL

(Web-Page) (Album / Profile)
leonard.wilson2009@hotmail.comShop the Official Coca-Cola Store!Click here for the Best Buy Free Shipping Offers
ArabicChinese (Simplified)Chinese (Traditional)DeutchEspanolFrenchItalianJapaneseKoreanPortugueseRussian
Personalized MY M&M'S® Candies

No comments: