Macintosh statistical software

SAS JMP 8 for the Macintosh review

[See JMP 10] JMP started out many years ago as John’s Macintosh Project, bringing visual exploratory statistics to the microcomputer. The program was puzzling; it replaced the Mac desktop with a program desktop, meant to bring a new user interface to statistics, along with the kind of three-dimensional exploratory graphing most people had only dreamt of.

Today, SAS-JMP 8 has a more conventional interface, but it is still advanced and designed for exploratory research. That is not to say it will not do most of what statistical researchers do — you can have a proper pre-determined research plan, and carry it out easily with JMP. By the same token, though, you can “play with your data” easily, with numerous shortcuts to make quick changes to the output or statistics.

Mac prefs

The writers have kept it Mac-friendly, to the point of having Mac-only preferences. Performance in our initial tests was blindingly fast, with instant response times, and no hint that this program is published by a company that only makes a single Mac-compatible product. It was, to say the least, amazing.

We even wrote, “There is no point in comparing JMP with PASW 18 (SPSS) in terms of speed; they are in different classes. Think about a Dodge Viper and a Toyota Yaris automatic and you may get the picture.”

Try to open a spreadsheet with four variables and 30,000 15-digit entries, and while SPSS PASW is busily digesting the idea, JMP has already opened it and displayed descriptive statistics... in an easily copied spreadsheet. Opening a small survey with 50 variables and a couple of hundred responses provides immediate gratification; scripts and search/replace operations are practically instant, too.

The JMP session begins with launching the program. At that point, the user is faced with two windows (both of which can be disabled from the preferences for a slightly faster launch); one provides basic starting commands for JMP, the other provides the “tip of the day.” 

starter

One interesting aspect of the “Starter” is that it’s not just, as it appears on first glance, the standard “here’s a bunch of ways to open new or old files” screen. On the left is a second menu of options; so you can actually use the Starter to run tests, too, if you want. It’s not a bad way for beginners or casual users to get used to the program, and it uses relatively jargon-free language along with icons to make the results clear.

models

You can open files from a shared database via ODBC, using a surprisingly versatile and user-friendly dialogue, or from the usual tab-delimited and CSV text files; the program does a good job of guessing what format the data is in, and unlike PASW/SPSS, does not force the user to select the data type before showing a list of options. Import from Excel is easy, and JMP is “Excel-savvy” enough to know about the different tabs that can be in a worksheet. The file open/save dialogues are standard Mac versions, so special folders (like the Desktop) appear where you’d expect, rather than being ignored (e.g. GraphicConverter).

Playing with data

Variables with long labels are problematic in PASW dialogue boxes; you can resize the box, but the variable lists only resize in proportion to the overall box, so you can end up with a dialogue box that fills the screen but doesn't show a full variable label. JMP, in contrast, lets you resize the each part of a dialogue box individually, so you can see a long variable label easily.

Changing variable names “the regular way” was clunky in each program. In PASW, one must manually go into the variable name, then type in the replacement (or use syntax); in JMP, one must double-click the variable name, then type in the replacement (or use syntax). But both programs let you select groups of variable names and deal with them in other programs. In JMP, you can swipe-select a group of variable labels from the left-hand list-window, copy, paste them into a spreadsheet, make changes there, and copy them back in again.

As with PASW 18, you can define each variable’s role in life, or leave them blank.

JMP 9 adds (wording from JMP):

Tabulation: JMP as Crystal Reports

The survey guy’s life is full of tabulations; and most tools out there make it somewhat clunky at best, including PASW 18’s setup (the module, incidentally, costs as much as all of JMP). The Tabulate command in JMP (Tables menu) allowed us to quickly set up these tables, after some reference to the liberally illustrated help function:

JMP software for Mac

Two things are immediately noticeable. One is the fact that we had two dialogue boxes open at once. You can play quite a bit in JMP with all sorts of windows open - dialogue boxes, input tables, output tables (which can be used as input tables), Help, etc. The main limit seems to be your monitor.

To get the table, we assigned value labels to a bunch of variables, then selected Tabulate, and dragged the variables up into the column, setting % of Total as the statistic. It took us a long time to figure it out; the main block was actually opening up the manual, which made the process rather obvious.

We like syntax, and discovered that while JMP doesn't make a big deal about it, it does have a good syntax facility, which is reached by going to the burgundy-colored triangle in an output section and selecting the Script submenu. This provides numerous options — adding a script to the journal, opening it in a new window, saving it with the data, etc. You can also re-run an analysis (say, after making some changes to the data) or re-open the control panel to restart it. It’s a very nice, flexible setup.

Here’s the script. Keep in mind the variable labels are also the variable names, at least in this file.

Tabulate(
Show Test Build Panel( 1 ),
Include missing for grouping columns( 1 ),
Add Table(
Column Table(
Columns by Categories(
:These changes will make us more productive.,
:These changes will make us happier.,
:I will play with JMP all day.,
:I will play with PASW all day.,
),
Statistics( Name( "% of Total" ) )
)
)
)

A stacked version:

stacked

Unique JMP features

Numeric output of JMP in our initial comparison tests (which used 30,000 randomly generated 15-digit numbers) was identical in PASW, Stata, JMP, and the free Megastat, indicating that all were coming up with the same numbers. Timing the tests was another story; JMP finished before we could look at the stopwatch. A regression of that dataset was instantaneous, including multiple multiple, high-resolution plots we could copy directly into Photoshop, summary of fit, analysis of variance, parameter estimates, residual plots, actual by predicted, etc.

From the output window, we could not only copy and paste tables or plots, but could change parameters or run additional tests from convenient submenus within the output.

It’s hard to beat JMP’s stunning and flexible graphics, its solid package of statistics, its helpful help which goes into both statistical issues and program issues, and its sheer responsiveness — not to mention the ease of taking its output and putting it into other software. For graphics, you can even have it use Photoshop file format (PSD) when copying.

One of the nice features of JMP is the ability to customize easily; unlike PASW (SPSS), you have a lot of control over the output. Also unlike PASW, of course, is the ability to simply copy and paste what you want out of the output window, without having to double-click to get into an editor, rearrange things, then leave the editor, then copy... or copy whole chunks of output at once.

prefs by chart

Glitches

We started to have some issues with a large production file exported in SAS format from SPSS, having 6,000 cases and 107 variables, including string variables. JMP seemed to stall out, using almost no CPU resources (according to Activity Monitor) but simply not moving. We could access the menus, but the spreadsheet view could not be changed, and trying to generate tables stalled out in the dialogue box.

We tried saving the file, quitting, and re-opening JMP to see if the problem was using up memory. Now, the spreadsheet view could not be opened at all, and again, trying to generate tables stalled the program. JMP showed no activity to speak of in the Activity Monitor, with 8 threads, while it sat motionless, not redrawing the screen. After a while, we figured out that the problem was the variable labels exported from SPSS. The original file, saved from SPSS into Excel, devoid of value and variable labels, brought JMP back to instantaneous results. So did other files, with truncated variable labels.

When you export from PASW / SPSS, depending on the file version you export to, variable labels can be truncated. PASW/SPSS users may also be annoyed by the treatment of missing values — they can be specified, but the change appears to be permanent, with the missing values removed.

While PASW 18 can read Stata and SAS files, SAS-JMP cannot read SPSS (PASW) files or newer SAS (7+) files. It can read Minitab files, SAS transport files, and the Linux version (but not Mac or Windows) can read OpenOffice files.

Because SPSS shares its data SDK, some programs natively export SPSS (PASW) data complete with data definitions (variable and value labels). While SAS itself can read SPSS and Stata files, JMP cannot. Exports in tab- or comma-delimited format lose their value and variable labels.

When saving SAS files from PASW, one is offered a handy checkbox of preserving value labels in a separate SAS syntax file. PASW value labels — often essential for avoiding mistakes — are, unfortunately, not readable in JMP, which tries but tosses out arcane errors:

uh-oh

Variable labels fare better, though it’s not a single-step process:

variable labels

Version 8.02 came out on October 30. Our copy’s "check for upgrades" told us we had the most recent version. We went to the upgrade site to download the updater, but there are several different versions of JMP, and the instructions on how to tell which one we had ... did not work.

K.J. Khoo added:

If a variable is pre-assigned a weight role, it triggers an error when an attempt is made to "recall" an analysis. SAS is aware of it, but instead of fixing it with an update, instead requires the user to upgrade to JMP 9. This bug was not present in JMP 7.

JMP makes a hash of weights in the fit model platform. This is perhaps most clearly seen by repeating the analysis using standardised weights. In effect, it treats weights as frequencies, although it provides for a pre-assigned role of frequency.

Conclusion: JMP for Macintosh is worth investigation

JMP is not just cheaper than PASW (for a standard license); no extra modules are needed, which encourages the user to branch out into new and different areas of statistics, modelling, and experimentation (though users are cautioned to fully understand the methods they’re using — the assumptions being made, for example — so spurious results aren’t acted upon as though they were real).

JMP has grown substantially in its range of capabilities, and, while it’s not always easy to translate “the SPSS way of doing things” over to JMP, it can do a surprisingly large range of statistical methods, and many people may find that they no longer need PASW once they have JMP; but the reverse is not as easily said, since JMP’s graphics are so extensive and more easily used. The hidden-syntax system works, and is quite clever, though it takes getting used to.

We think we will be exploring JMP for a long time to come — and discovering new capabilities all the time.

[See JMP 10]JMP Software (Company Site)

The original quick review: notes from the MacStats Home Page

I’m testing SAS-JMP 8, and I am very impressed. The writers have kept the Mac-friendliness alive, with a set of Mac-only preferences, blindingly fast performance, instant response times, and no hint that this program is published by a company that only makes a single Mac-compatible product. It is, to say the least, amazing.

Comparisons to SPSS cannot be favorable... at least in speed. Try to open a spreadsheet with four variables and 30,000 entries, and while SPSS PASW is busily digesting the idea, JMP has already opened it and displayed descriptive statistics... in an easily copied spreadsheet. Run the same descriptives in PASW, and you get a table with a bunch of dots — because the columns are the wrong size! This is a stunner.

Numeric output of JMP in our tests (which used 30,000 randomly generated 15-digit numbers) was identical in PASW, Stata, JMP, and the free Megastat. In this group, JMP was by far the fastest (finished before we could look at the stopwatch), followed closely by PASW and Stata (around 1 second each), and then by Megastat, which at 3 seconds was still quite reasonable.

Regressions in JMP were instantaneous. That includes multiple multiple, high-resolution plots we could copy directly into Photoshop, summary of fit, analysis of variance, parameter estimates, residual plots, actual by predicted, etc.

From the output window, we could not only copy and paste tables or plots, but could change parameters or run additional tests from convenient submenus within the output.

It’s hard to beat JMP’s stunning and flexible graphics, its solid package of statistics, its helpful help which goes into both statistical issues and program issues, and its sheer responsiveness — not to mention the ease of taking its output and putting it into other software. For graphics, you can even have it use Photoshop file format (PSD) when copying.

We’ll be posting details and screen shots this summer, in a dedicated JMP review.

Sponsored by Toolpack Consulting

Toolpack: for information-based organizational change

Copyright © 2005-2024 Zatz LLC. All rights reserved. Created in 1996 by Dr. Joel West; maintained since 2005 by Dr. David Zatz. Contact us. Terms/Privacy. Books by the MacStats maintainer