Vous êtes sur la page 1sur 20

The JMP Advantage

Lee Creighton

Bradley Jones

John Sall

Annie Zangi
Table of Contents

Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

User Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Integrated Graphics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Interactivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Problem-centric vs. Tool-centric Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Scripting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Six Sigma® Quality Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6


Design of Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Control Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Attribute Gage Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Gage R&R And Variability Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Statistical Modeling Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10


Fitting Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Mean-dispersion Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Nonlinear Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Neural Nets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Recursive Partitioning (Trees) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
The JMP® Advantage

Executive Summary
JMP® provides comprehensive statistics and an interactive environment for designing
experiments, analyzing and exploring data, and uncovering trends that might otherwise go
unnoticed. JMP can be customized and tailored to fit the varied needs and levels of practitio-
ners in your organization, and its graphical interactivity enables everyone to make a contribution
to the productivity gains promised by Six Sigma. JMP is ideally suited for those needing a desk-
top statistics package that is suited for all users, including every level of Six Sigma practitioners
from Master Black Belts on down.

This paper highlights the distinctive aspects of JMP, its user interface, Six Sigma tools and
advanced statistical modeling capabilities.

1
The JMP® Advantage

User Interface
Integrated Graphics
Every JMP® analysis has a graph as its central
element and often has a graph for every sta-
tistical test. JMP fluidly scrolls the graphs and
reports together as a single document in single
window. JMP has innovative graphs for some
analyses that help you interpret the analysis,
find patterns, find points that don’t fit patterns,
and follow up on any pattern, outlier, or analyti-
cal insight you have. JMP automatically gives
graphs, rather than you having to ask for them.

JMP interactive profilers promote exploring


opportunity spaces. In fitting equations to data,
the fitting is only half the job. Interpreting the fit,
understanding the fitted response surface, and
finding factor values to optimize the responses
is desirable.

JMP displays graphs that help users interpret


Prediction and optimization are integrated into analyses and discover patterns in data..
JMP’s fitting platforms through the use of three
platforms: the prediction profiler, which shows
response surface slices through each factor;
the contour profiler, which shows horizontal
response surface slices with respect to two
factors at a time, and the surface profiler, which
shows a 3-D rendered surface. All of these are
completely interactive.

In the case of the Prediction Profiler, not only


can you interactively explore the different slices
of the response surface, but you can specify
the desirability of levels of each factor and have
JMP find optimal factor settings.

Most platforms support commands to save


predicted values or prediction formulas.
You therefore have easy access to the
mathematical model.

Prediction, contour, and surface profilers help


users interpret the fit and find factor values to
optimize the response.

2
The JMP® Advantage

Interactivity
JMP’s analyses are interactive, not static. In JMP, commands are available interactively to
extend an analysis or provide more details. As an analysis unfolds, further opportunities appear.
You don’t have to think ahead and specify everything you want before you launch the analysis.

JMP encourages active exploration. Because each analysis is interactive, you are naturally
encouraged to follow clues to dig deeper into the analysis. You don’t have to relaunch the analy-
sis to find available options. If you turn on an option that doesn’t contribute to your understand-
ing, you can turn it back off, uncluttering the report.

JMP features dynamic linking. It is common


to identify a point in one graph and want to see
where that point falls in other graphs. This is
easy in JMP—just click on the point.

The rows in JMP data tables are dynamically


linked. Click on a histogram bar, and all the
rows for that bar become highlighted, leading
to highlighting the points in any open graph or
analysis. Once highlighted, you can adjust color,
marker, and label styles, as well as exclude
them from future analyses—all by shifting your
finger from the left mouse button used to high-
light, to the right mouse button to access a con-
textual command. Click on data point, and all the rows for that point
become highlighted in any open graph or analysis.

JMP reports are interactive.


You see a report table and
want to sort it by the p-value?
Easy—just right click to the sort
by column command. You want
to make that table into a data
table? Again, just right click. You
want to rearrange the columns
of the table? Just drag with the
hand cursor. Want to see more
(or fewer) decimal places in a
report column? Just double click
and enter the number you want.

In addition, all graphs can be


dragged to any size. Related
graphs resize. Drag axes to JMP lets you select analyses that are
change them, expand them, contract them. appropriate to the situation.

Right click to do many other customizations.

3
The JMP® Advantage

Problem-centric vs. Tool-centric Design


JMP’s interface is problem-centric. JMP’s analyses are organized into platforms. Each plat-
form is a launch point for the analysis of a statistical situation. The applicable statistical tools are
automatically applied and new analysis options are revealed based on the types of variables in
the analysis. This allows for a short menu organization that can be mastered quickly. In addition,
novice users are not burdened with having to know the appropriate tool for every special case in
every statistical problem. By arranging menus as simply a list of tools, other software packages
can place this burden squarely on the shoulders of the user.

With JMP, the launching process is often just specifying the variables you want, rather than all of
the details of the analysis. In addition, analysis dialogs are non-modal. You are free to work on
other windows and check details in deciding what you want before you commit or dismiss a
dialog.

With JMP, you select the data table and the type of analysis you want, drag and drop variables, specify
parameters, and run the analyses.

Data Table
JMP data is richer. JMP’s data tables support a rich set of metadata, including table name-
value pairs, scripts that you can store with the data, many types of property data, including
coding, value labels, models, and so on. In JMP, you identify a variable as continuous
or categorical once, and the setting is remembered and doesn’t need to be respecified in
each analysis.

JMP supports active and persistent formulas. JMP has column formulas which stay with the
data, and are evaluated as data is changed or added. Using a formula, you can easily fill a col-
umn with data.

4
The JMP® Advantage

Scripting
Early JMP adopters wanted to use extend their use of JMP in three ways. They wanted to
• Automate common analysis tasks
• Extend JMP’s capabilities by adding their own statistical analyses
• Create animations to demonstrate statistical points.

To answer these requests, JMP now includes a complete scripting language. In fact, JMP can
produce scripts itself that mirror actions you perform interactively, allowing you to quickly use
JMP to monitor frequently updated processes.

JMP captures your analyses as a script you can replay or edit. In this script to demonstrate how Kernel
densities work, a handle allows users to interactively change the bandwidth while the script is running.

5
The JMP® Advantage

Six Sigma® Quality Tools


Design of Experiments
The majority of experiments done in industry today use two-level fractional factorial designs.
Using these designs often involves changing the problem to fit the requirements of the design.
For instance, many real-world systems have categorical factors with more than two levels. To
use a traditional two-level factorial design involves ignoring all but two of the possible settings.

By contrast, JMP’s Custom designer builds a design to match the requirements of the problem.
The Custom designer is a general-purpose tool supporting screening, RSM, and mixture experi-
ments. By supporting these objectives with a single designer, JMP returns the work to being
problem-centered, rather than tool-centered.

The Custom designer is also flexible, allowing


• Any number of factors
• Any number of levels of categorical factors
• Blocking factors with any number of runs per block
• Arbitrary constraints on the settings of factors
• Fixed covariate factors
• Mixture factors
• Any combination of the above factor types

Pre-formulated designs require specific sample sizes. For example, two-level factorial designs
have sample sizes that are powers of two. The Custom designer can build designs for any
sample size greater than or equal to the number of estimated parameters. This can result in
substantial savings of resources and time.

Optimal experiments involving dozens of factors are infeasible for older algorithms that required
a candidate set (that grows exponentially with the number of factors). The Custom designer
generates either D-optimal or I-optimal designs using the coordinate exchange algorithm.
Because the coordinate exchange algorithm does not require a candidate set, the Custom
designer can quickly produce designs for these large problems.

There are situations where there are dozens of factors, but resource constraints limit the
number of runs to be fewer than the number of factors. The Custom designer can generate
supersaturated designs using an extension of Bayesian D-optimality that allows these
situations to be analyzed.

Before running the experiment, the JMP prediction variance profiler lets you see the prediction
variance relative to the noise variance to determine if the experiment has enough predictive
specificity.

6
The JMP® Advantage

Before running an experiment, you can use the


prediction variance profiler to help determine if the
experiment has enough predictive specificity.

Augmenting existing designs. The Augment designer can add a specified number of runs
either D- or I-optimally to any existing experiment. This addition can occur simultaneously with
new factor ranges and a more complex model specification. Thus, the Augment designer sup-
ports a process of sequential experimentation that speeds the discovery of optimal settings.

Space-filling designs. Many leading-edge companies employ computer models to prototype


design ideas, saving the expense of building physical prototypes. These computer models are
typically complex and involve many variable parameters. Space-filling designs are useful for
building simpler empirical approximations to the computer models. An engineer can then use
JMP’s graphical profilers to visualize the results, quickly determine the dominant variable rela-
tionships, relative sensitivities, and optimal settings within the parameter ranges.

Control Charts
JMP produces all common control charts. In addi-
tion, it can capture data in real time (for an instrument
attached to a serial port) and augment a chart instantly.
Tests for out-of-control points (Western Electric rules
and Westgard Rules) can trigger an email, speak a
warning, or execute a script.

Control charts offer a variety of ways


to analyze and monitor process data
that has been captured in real time or
stored in data tables.

7
The JMP® Advantage

Attribute Gage Charts


Attribute gage charts in JMP have live graphs. JMP gives summary graphs on both the parts,
or any variety of x variables given and a summary graph showing the average agreement on
the raters. Because the graphs are live, you can click on points for further exploratory analysis.
Below you see the data from the AIAG MSA 3rd edition on Attribute Gage charts. You can easily
see the problem parts.

JMP Attribute Gage Charts are live llinked for additional exploration of variables.

Agreement counts. In addition to Kappa statistics, JMP gives agreement reports with counts
showing the number correct with each level of the response for each rater, by name.

For some companies, parts arriving off an assembly line which are
incorrectly labeled as “Working” by the inspector are more of a problem
than those labeled “Not Working”

Effectiveness summary reports show total correct/total possible


responses, each summarized by rater to provide an overall summary
together with confidence intervals.

8
The JMP® Advantage

The Misclassification table. The Misclassification table is particularly useful when you have
more than two possible responses and you’d like to track the misclassifications.

In the call data, you can see that 8 calls were incorrectly sent to the DR group, when they should have
been routed to the DT group.

Gage R&R And Variability Charts


JMP’s variability charts are not limited to only two x variables, but can handle several. For exam-
ple, the graph shown here is from data that has variables for Part, Operator and Instrument.

The platform gives output including graphs and variance components. JMP is only limited by the total
number of levels (232-1).

REML available in variability charts. When data is unbalanced, JMP automatically computes
variance components using REML.

9
The JMP® Advantage

Statistical Modeling Capabilities


Fitting Models
JMP uses a general problem-solving approach rather than employing a larger collection of more
specific tools. When you have a combination of specific needs, the toolbox approach, used
in other statistical software programs, may not have an entry corresponding to your need. For
example, some packages claim to support random effects and variance components estimation
but that support is for categorical effects and balanced data. Some packages limit you to using
a simple method of moments on expected mean squares. JMP, on the other hand, implements
random effects in the general case, allowing unbalanced data, continuous regressors, random
regressors, and uses the REML method.

JMP fits models using the REML method and implements random effects in the general case to allow for
any type of effects, unbalanced data, continuous regressors and random regressors.

In JMP there is one main linear fitting facility that has almost all the features available with any
model. The same facility that supports random effects supports cube plots and multiple com-
parisons and dozens of other features upon request. Learn this single dialog to launch stepwise,
MANOVA, log-variance, nominal logistic, ordinal logistic, parametric survival, and proportional
hazards survival models. You do not have to learn different dialogs to launch models with differ-
ent features.

10
The JMP® Advantage

Suppose that you have five hard-to-change factors in a fractional factorial whole plot design,
subdivided by two other factors in the split plots. In JMP, you simply specify a high-order
interaction of the whole plot factors as a random effect to absorb the whole plot error, and
everything else is automatic. JMP treats random effect interactions sparsely, making this a rea-
sonable model, matching SAS PROC MIXED.

JMP lets you include fractional factorial whole plots in a split plot analysis.

11
The JMP® Advantage

Mean-dispersion Models
Sometimes, different factor settings lead to changes in both the mean and variance of the
response. JMP supports a way to model this situation. JMP does a series of EM alternations
of two linear models, one for the mean response, the other for the log of the variance of the
response, and produces maximum likelihood estimates of the models.

Setting Htime = -1 leads to dramatically reduced variability. JMP supports mean dispersion models with its
other model diagnostic tools, like the prediction profiler.

12
The JMP® Advantage

Nonlinear Regression
JMP has a platform for nonlinear modeling that allows you to build your model by entering
a formula or choosing from a model library. It is a general-purpose facility that is interactive,
produces its own analytic derivatives, use loss functions to do maximum likelihood, calculate
profile-likelihood confidence limits, and more. In fact, in most platforms that offer likelihood-
based estimates, JMP offers profile-likelihood confidence limits, which are much more reliable
than the normal-asymptotic limits. In addition to nonlinear regression, you see these limits in
logistic regression, ordinal logistic regression, parametric survival models, and survival
regression models.

Neural Nets
Often, quadratic models are not adequate to capture the curvature in a response surface.
JMP offers neural nets, a more flexible tool for modeling curvature. This is especially valuable
when there are multiple responses, since neural net models are more parsimonious. JMP pro-
vides controls to tune the behavior of the neural net algorithm, allowing you to match the fit to
your problem.

As with other modeling platforms, Neural Nets have a complete set of tools to explore and
optimize the model.

13
The JMP® Advantage

Recursive Partitioning (Trees)


The key to a Six Sigma® discovery may be to find the important relationships in a set of data.
Exploring for relationships is the idea behind a Multi-vari study. Ordinary statistics and graphs
can find these relationships if you already know where to look. If you don’t know where to look,
and you have a lot of places to look, you need something with more power in finding relation-
ships—a data mining tool. Recursive partitioning (trees) is the preeminent tool that can look
through hundreds of columns, and for each column hundreds of cut points or classification
arrangements, and report on what is most strongly associated with the response. For each
split, other relationships become apparent, just suggesting a recursive splitting of the data at
many steps until you have characterized which conditions are associated with good and with
bad results.

These are the clues that can lead to breakthrough discoveries and changes. JMP has a particu-
larly interactive and graphical implementation of recursive partitioning that is easy to use, fast
and powerful up to hundreds of columns, and hundreds of thousands of rows.

JMP presents recursive trees in an interactive, graphical report that gives you control over splitting and
pruning the tree’s nodes.

14
The JMP® Advantage

Conclusion
JMP was written from scratch with modern graphical interactivity in mind. JMP has made huge
investments in features that matter to engineers and researchers: DOE, modeling, surface
exploration, data exploration, statistical discovery. We have kept our product focused, friendly,
and engaging for the user.

15
JMP Sales JMP Japan
SAS Campus Drive Inui Bldg Kachidoki
Cary, NC 27513 1-13-1 Kachidoki
US: 877-59-GOJMP Chuo-ku Tokyo 104-0054 Japan
International: (1) 919-677-8000 Tel: (81) 3 3533 3887
Fax: (1) 919-677-4444 Fax: (81) 3 3533 1600
jmpsales@sas.com jmpjapan@sas.com
www.jmp.com www.jmp.com/japan
SAS, JMP, and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and
other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies. Copyright © 2005,
All rights reserved. Six Sigma is a registered trademark of Motorola, Inc. 349306US_0805

Vous aimerez peut-être aussi