Massachusetts Institute of Technology



What’s New in

JMP® Version 3.2

February 1997

This document gives a detailed description with examples of all changes and enhancements that have occurred after JMP version 3.1. It assumes you have access to the manuals that came with the full JMP software package.

The following three manuals are included with the full JMP package:

• The JMP Introductory Guide is a collection of tutorials designed to help you learn JMP strategies. The JMP tutorials range from single-step procedures to complex analyses. You can read the tutorials for reference, or work through them step by- step. Each tutorial uses a file from the sample data folder. By following these examples, you can quickly become familiar with JMP menus, graphical displays, options, and report windows.

• The JMP User’s Guide has complete documentation of all JMP menus, an explanation of data table manipulation, and a description of the calculator. There are chapters that show how to do common tasks such as manipulating files, transforming data table columns, and cutting and pasting JMP data, statistical text reports, and graphical displays

• The JMP Statistics and Graphics Guide documents the statistical platforms, discusses statistical methods, and describes all report windows and options.

JMP® Version 3.2

© Copyright 1997 by SAS Institute Inc., Cary, NC, USA

All rights reserved. Printed in the United States of America. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, or otherwise, without prior written permission of the publisher, SAS Institute Inc.

Information in this document is subject to change without notice. The software described in this document is furnished under the license agreement printed on the envelope that contains the software diskettes. The software may be used or copied only in accordance with the terms of the agreement. It is against the law to copy the software on any medium except as specifically allowed in the license agreement.

First printing, November 1996

JMP®, JMP Serve®, and SAS®, are registered trademarks of SAS Institute Inc. All trademarks above are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries.

® indicates USA registration.

Contents

Welcome to JMP 1

System Hardware and Software Requirements 2

Windows 2

Macintosh 2

Before you Install 3

Register Your Product 3

Installing JMP 4

Windows Installation 4

Macintosh Installation 4

Customer Support 4

Sales Support 4

Training and Education 4

Technical Support 4

JMP Information on the World Wide Web 5

Overview of Enhancements in JMP Version 3.2 6

General 6

Analysis Windows 6

Data Tables - The Tables menu 6

Design of Experiments 7

Calculator 7

Analyze Menu - Distribution of Y: Continuous Variables 8

Analyze Menu - Distribution of Y: Nominal or Ordinal Variables 8

Fit Y by X - general 8

Fit Y by X - Continuous by Continuous 8

Fit Y by X - Continuous by Nominal 9

Fit Y by X - Nominal by Nominal 9

Analyze Menu - Fit Model Dialog 9

Analyze Menu - Fit Model : Standard Least Squares 9

Analyze Menu - Fit Model: Effect Screening 9

Analyze Menu - Fit Nonlinear 11

Analyze Menu - Correlation of Y’s 11

Analyze Menu - Cluster 11

Graph Menu - Control Charts 11

Tools Menu 12

Windows Menu 12

Calculator - Random Number Functions 12

Calculator - Probability Functions 15

Analyze Menu - Distribution of Y: Continuous Variables 17

More Moments 17

Stem and Leaf Plot 18

Test Mean=value 18

Outlier Box Plot 19

Capability Analysis 19

Normal Quantile Plot 19

Analyze Menu - Distribution of Y: Nominal/Ordinal Variables 21

Test Probabilities for Nominal Distributions 21

Popup options for Frequency Table 21

Analyze Menu - Fit Y by X: Continuous by Continuous 22

Nonpar Density 22

Fit Transformed 22

Other Changes to the Fit Y by X Platform for Continuous Y and X 23

Analyze Menu - Fit Y by X: Continuous by Nominal 24

Normal Quantile Plot 24

The Matching Variables Command 24

Means Diamonds Have 95% Overlap marks 26

Other Changes to the Fit Y by X for Continuous Y and X 26

Analyze Menu - Fit Y by X: Nominal/X by Nominal Y 27

Cochran-Mantel-Haenszel Test 27

Analyze Menu - Fit Model 28

Standard Least Squares Parameter Estimates Details 28

Screening: Prediction Profile 28

Screening: Effect Screening Option for Coded Estimates 29

Screening: Contour Profiler 31

Analyze Menu - Fit Nonlinear 32

Analyze Menu - Correlation of Y’s 33

Analyze Menu - Cluster 33

Graph Menu - Control Charts 33

Variability Analysis (Gage R&R Charts) 33

Gage R&R Charts 35

Gage R&R Variability Report 35

[pic]

Welcome to JMP

JMP® is statistical discovery software that can help you explore data, fit models, discover patterns, and discover points that don’t fit patterns. As statistical discovery software, the emphasis in JMP is to interactively work on data to find out things.

• Using graphics, you are more likely to make discoveries. You are also more likely to understand the results.

• With interactivity, you are encouraged to dig deeper, for one analysis can lead to a refinement, one discovery can lead to another discovery; and you can experiment with statistics to improve your chances of discovering something important.

• With a progressive structure, you build context that maintains a live analysis, so you don’t have to redo analyses, so that details come to attention at the right time.

The job of software is to create a virtual workplace. The software has facilities and platforms where the tools are located and the work is performed. JMP provides the workplace that we think best for the job of analyzing data. With the right software workplace, researchers will celebrate computers and statistics, rather than avoid them.

JMP aims to present a graph beside every statistic. You can and should always see the analysis in both ways, statistical text and graphics, without having to ask for it. The text and graphs stay together.

JMP is controlled largely through point and click, mouse manipulation. If you click on a point in a plot, JMP identifies and highlights the point in the plot, also highlights the point in the data table, and highlights it everywhere else the point is represented.

JMP has a progressive organization. You begin with a simple surface at the top, but as you analyze you see more and more depth. The analysis is alive, and as you dig deeper into the data, more and more options are offered according to the context of the analysis.

In JMP, completeness is not measured by the “feature” count, but by the range of applications, and the orthogonality of the tools. In JMP, you get more feeling of being in control in spite of less awareness of the control surface. In JMP you get a feeling that statistics is an orderly discipline that makes sense, rather than an unorganized collection of methods.

The statistical software package is often the point of entry into the practice of statistics. JMP endeavors to offer fulfillment rather than frustration, and empowerment rather than intimidation.

[pic]

System Hardware and Software Requirements

Windows

• Operating System

Microsoft Windows NT 3.51 or greater

Microsoft Windows 95

MS-DOS 5.0 or later with Microsoft Windows 3.1

• Machines Supported

IBM-compatible personal computer with a 386 or higher microprocessor

• Math Coprocessor

A math coprocessor is not required

• Distribution Media

Diskettes

• Memory

8 MB minimum with 8 MB of swap file space

16 MB recommended with 10 MB of swap file space

Note: The memory requirements are in addition to the amount of memory needed to run the operating system.

• Hard Disk Storage

6 MB minimum (without Win32s)

8 MB minimum (with Win32s)

8 MB recommended (without Win32s)

10 MB recommended (with Win32s)

Macintosh

• Operating System

System 7.0 or greater

• Machines Supported

68k Macintosh: JMP runs on any MacPlus or later Macintosh model with a 680x0 processor.

PowerPC Macintosh: JMP runs on any PowerPC Macintosh processor

• Math Coprocessor

A math coprocessor is not required.

• Distribution Media:

Diskettes

• Memory

4 MB minimum

8 MB recommended

Note: The memory requirements are in addition to the amount of memory needed to run the operating system.

• Hard Disk Storage

4 MB minimum

8 MB recommended

Before you Install

Before you install JMP be sure to make backup copies of your JMP diskettes.

To protect your software, lock the JMP installation disks by sliding the locking tab on each disk so that the hole in the disk is open. Then make backup copies of the disks before you begin the installation process. If you need additional information on locking or copying disks, see the documentation that came with your Macintosh or Windows computer

Register Your Product

If you are a new user please take a moment to complete and mail the postage paid product registration card for your new software. Keep your portion of the card with your serial number in a safe easily accessible place. As a registered JMP owner you are eligible for:

• free technical support

• software updates

• JMPer Cable, JMP’s quarterly newsletter

• a free JMP T-shirt or JMP mug

[pic]

Installing JMP

The JMP package contains program disks that have installation instructions printed on the disks label.

Windows Installation

To install JMP under Windows insert “Disk 1” and double-click on the INSTALL.EXE file. The Windows version of JMP runs best on Windows 95 or Windows NT. When run on Windows 3.1 a special system library called Win32s is required to allow for the 32-bit application. If the Win32s library (version 1.25 or later) has not been installed, you must run the Win32s setup provided in your JMP package before installing JMP software.

Macintosh Installation

To install JMP on a Macintosh, insert “Disk 1” and double-click on the JMP Installer. Follow the instructions to complete a standard install. We recommend that you do a standard install the first time you install JMP.

[pic]

Customer Support

Sales Support

Sales support for JMP is available by calling SAS Institute at 919-677-8000 and ask for JMP sales. Hours of operation are 9:00 am to 5:00 pm Eastern Standard Time.

Training and Education

JMP Training and Education information is available by calling SAS Institute at

919 677-8000 x7312 during hours of operation, 9:00 am to 5:00 pm Eastern Standard Time.

Technical Support

Technical support is provided for as long as you license the software (annual license user) or for one year after purchasing the software provided you have returned your registration card (you are a registered user) and continues when you purchase upgrades. JMP technical support is divided into JMP statistical support and JMP system support (nonstatistical).

If you encounter errors or have questions regarding the installation, you may contact Technical Support via the World Wide Web, telephone, fax, dial-up computer access, or e-mail.

• For technical support via the World Wide Web, use the following URL:



• For technical support by phone, call

(919) 677-8008 or (919) 677-8118 between 9 a.m. and 5 p.m., Eastern Standard time, on SAS Institute business days.

• The fax number for technical support is

(919) 677-4444.

• Electronic mail support is available through the Electronic Mail Interface to Technical Support(EMITS). This facility allows you to track a technical support problem or add information to a previously reported problem via email. To obtain more information on EMITS, send electronic mail to SUPPORT@ with the body of the message containing the command:

help

JMP Information on the World Wide Web

All the information in this section is also available on the world wide web at the URL:



This web page also gives you the ability to

• download a free JMP demo

• download current patches

• see a technical overview of JMP statistics

• see a description of all training courses available

• find out more about purchase and license options

[pic]

Overview of Enhancements in JMP Version 3.2

This overview lists changes and enhancements between JMP version 3.1 and version 3.2. Following sections describe most of these features in more detail and give examples.

General

• This version of JMP for Windows contains an experimental real-time data capture feature. This feature is only available when JMP is running under Windows 95 and Windows NT 3.51 or higher. If you have an instrument connected to one of the COM ports of your computer and create a new column in your spreadsheet with its data source set to "Instrument", then JMP will attempt to read the data which is sent by your instrument to the COM port and enter it into the new column. In order for this feature to work properly, when you set the data source of the new column to "Instrument", you must correctly configure the COM port information in accordance with the properties of your particular instrument. This feature is experimental and is not guaranteed to work with all instruments.

• The Import Command can read FACS files. The FACS file import reads files from flow cytometery software packages that write their data files in the standard FCS format proposed by the International Society for Analytical Cytology. FCS versions 1,2, and 3 are supported. This includes most commercial Flow Cytometry systems from Becton Dickenson, Coulter Electronics, Ortho, Cytomation and others.

• The Open Command can read SAS data sets directly. SAS data sets are converted to JMP tables complete with column notes, table info, and conversion of SAS date/time values to JMP date/time values.

• You can now drag text files onto the JMP application and JMP automatically attempts to Import them as text.

• There is a new preference for gray backgrounds in analysis windows.

Analysis Windows

• When you COMMAND-click on a selected point in a plot, the point is deselected.

• p-values show as “|t|” is the significance level or p-value associated with the

Orthog t-test values.

• The Scaled Estimate report column has been removed.

• The Pareto plot is now done with respect to the normalized estimates, which are more appropriate, especially when quadratic and crossed uncoded terms are present.

• An additional warning line appears for non-orthogonal designs stating “Each Orthog Estimate is conditioned on the effects before it.”

Screening Prediction Profiler

• In the Prediction Profiler, new dialogs appear when you OPTION-click the graph to allow you to change the current value, the scale, and the number of grid points for continuous factors. Being able to change the number of grid points makes much more accurate desirability settings possible.

• In the Prediction Profiler, the popup menu has four items:

1) Confidence Intervals, which are initially on

2) Desirability Functions

3) Most Desirable in Grid, which searches all possible levels and sets the factors at the most desirable settings

4) Reset Grid, which brings up a dialog for each continuous factor.

• In the Save $ menu, Grid of Predicted saves predictions for all combinations in the grid, including the desirability.

• If you want to use the Prediction Profiler, but want different models for different responses, you can do a separate fit with these responses, saving the predicted value, and using the predicted value rather than the original response in the profiler.

• The Interaction Plots are interactive with respect to the current values of the profiler.

• A new Contour profiler offers interactive contour plots of the response surface that are use-ful for multiple-response optimization. Optionally, the contour profiler can show mesh plots.

Screening Contour Profiler

The check-mark popup menu on the screening platform has the Contour Profiler option that brings up an interactive contour profiling facility. This is useful for optimizing response surfaces graphically.

Analyze Menu - Fit Nonlinear

• The nonlinear platform has two new save menu commands: Save Pred Confid and Save Indiv Confid for saving confidence intervals using the asymptotic linear approximation.

Analyze Menu - Correlation of Y’s

• There is an additional principal component command to use covariance matrices, or uncentered and unscaled.

Analyze Menu - Cluster

• The rows in the data table and in the dendrogram are colored and marked. The color advances in the dendrogram tree for as long as the colors agree. The color/marker option is now a state that forces immediate recoloring as the diamond control is moved. Also, you can use the brush tool on the dendrogram.

Graph Menu - Control Charts

• Control Charts can do variability charts with variance components.

• The Exclude row state is used when a control chart is first invoked. It is used in estimating sigma if it is not specified. Thus, all statistics which are based on the sigma (like the control limits) are affected. Subgroup statistics used in plotting the charts are not affected; that is, points based on Excluded rows will be displayed.

Tools Menu

The Annotate tool has enhanced features. After you create a test box for annotation (a yellow-stickie-type note), you can hold down the OPTION key (Mac) or ALT key (Windows) and click on the sticky to bring up a menu that lets you:

|• change the color of the note |Annotate Tool Popup Menu |

|• change the type of border around the note |[pic] |

|• specify no border | |

|This enables you to create notes that act as editable| |

|fields that can be used to change the axis labels of | |

|plots and charts. | |

Windows Menu

• If there is more than one data window open, the Close Data Windows command under the Window menu displays a dialog that gives you the option of saving changes to all modified tables (without prompt), discarding changes to all opened tables (without prompt), or taking a prompt for each modified table.

[pic]

Calculator - Random Number Functions

|[pic] |The calculator has the following new random number |

| |functions: |

| |exponential, triangular(midpoint), Cauchy, gamma(alpha), |

| |poisson(lambda), binomial(n,p), geometric(p), negative |

| |binomial(n,p), and random number seed = conditional |

| |assignment |

The random number functions in JMP appear in formulas preceded by a “?” to indicate randomness. Each time you click Evaluate these functions produce a new set of random numbers. The following is a description of all the random number functions:

Uniform

?uniform generates random numbers uniformly between 0 and 1. This means that any number between 0 and 1 is as likely to be generated as any other. The result is an approximately even distribution. You can shift the distribution and change its range with constants. For example, 5 + ?uniform •20 generates uniform random numbers between 5 and 25.

Normal

?normal generates random numbers that approximate a normal(0, 1) distribution. The normal distribution is bell shaped and symmetrical. You can modify the Normal function with constants to specify a normal distribution that has a different mean and standard deviation. For example, 5+?normal•2 generates a normal distribution with a mean of 5 and standard deviation of 2.

Exponential

?exponential generates a single parameter exponential distribution for the parameter lambda=1. You can modify the Exponential function to use a different lambda. For example, 0.1•?exponential-0.1 generates an exponential distribution for lambda=0.1. The exponential distribution is often used to model simple failure time data, where lambda is the failure rate.

Cauchy

?Cauchy generates a Cauchy distribution with location parameter 0 and scale parameter 1. The Cauchy distribution is bell shaped and symmetric but has heavier tails than the normal distribution. A Cauchy variate with location parameter alpha and scale parameter beta can be generated with the formula alpha+beta•?cauchy.

Gamma

?gamma(alpha [pic] ) gives a gamma distribution for the parameter, alpha, you enter as the function argument. The gamma distribution describes the time until the kth occurrence of an event. The gamma distribution can also have a scale parameter, beta. A gamma variate with shape parameter alpha and scale beta can be generated with the formula beta •?gamma(alpha). If 2•alpha is an integer, a chi-square variate with 2•alpha degrees of freedom is generated with the formula 2•?gamma(alpha).

Triangular

?triangular(mid [pic] ) generates a triangular distribution of numbers between 0 and 1, with the midpoint you enter as the function argument. You can add a constant to the function to shift the distribution, and multiply to change its span.

Shuffle

?shuffle selects a row number at random from the current data table. Each row number is selected only once. When Shuffle is used as a subscript, it returns a value selected at random from the column that serves as its argument. Each value from the original column is assigned only once as Shuffle’s result.

Poisson

?poisson(lambda [pic]) generates a Poisson variate based on the value of the parameter, lambda, you enter as the function argument. Lambda is often a rate of events occurring per unit time or unit of area. Lambda is both the mean and the variance of the Poisson distribution.

Binomial

?binomial( [pic] , probability [pic]) returns random numbers from a binomial distribution with parameters you enter as function arguments. The first argument is n, the number of trials in a binomial experiment. The second argument is p, the probability that the event of interest occurs. When n is 1, the binomial function generates a distribution of Bernoulli trials. For example, n=1 and p=.5, gives the distribution of tossing a fair coin. The mean of the binomial distribution is np, and variance is np(1–p).

Geometric

?geometric(probability [pic] ) returns random numbers from the geometric distribution with the parameter you enter as the function argument. The parameter, p, is the probability that a specific event occurs at any one trial. The number of trials until a specific event occurs for the first time is described by the geometric distribution. The mean of the geometric distribution is 1/p, and the variance is (1–p)/p2.

Negative Binomial

?negBinomial( [pic] , probability [pic] ) generates a negative binomial distribution for the parameters you enter as function arguments. The first parameter is the number of successes of interest (r) and the second argument is the probability of success (p). The random variable of interest is the number of failures that precede the rth success. In contrast to the binomial variate where the number of trials is fixed and the number of successes is variable, the negative binomial variate is for a fixed number of successes and a random number of trials. The mean of the negative binomial distribution is (r(1–p))/p and the variance is (r(1–p))/p2.

Random Number Seed

This function lets you start a random number sequence with a seed you specify. To use the Random Number Seed function, assign it a value using the Assignment function found in the Conditions functions, and use the random number function

|you want as the results clause of the Assignment function. The example|[pic] |

|shown here uses the number 1234567 as the seed to generate a sequence | |

|of uniform random numbers. | |

[pic]

Calculator - Probability Functions

|[pic] |Probability Functions now include the density function|

| |for the Normal distribution . Also, there are |

| |functions that calculate densities, quantiles and |

| |probabilities for the gamma and beta distributions: |

[pic]

The Normal Distribution function accepts a quantile argument from the range of values for the standard normal distribution with mean 0 and standard deviation 1 . It returns the probability that an observation from the standard normal distribution is less than or equal to the specified quantile. For example, the expression normDist(1.96) returns .975, the probability that an observation from the standard normal distribution is less than or equal to the 1.96th quantile. The Normal Distribution function is the inverse of the Normal Quantile function.

[pic]

|The Normal Density function accepts a quantile argument from |[pic] |

|the range of values for the standard normal distribution. It | |

|returns the value of the standard normal probability density | |

|function (pdf) for the argument. For example, you can create a | |

|column of quantile values (X) and use normDensity(X) to | |

|generate density values. Then use GraphÆOverlay to plot the | |

|normal density by X, as shown to the right. | |

[pic]

The Normal Quantile (probit) function accepts a probability argument p, and returns the pth quantile from the standard normal distribution. For example, the expression normQuant(0.975) returns the 97.5% quantile form the standard normal distribution, which evaluates as 1.96. The Normal Quantile function is the inverse of the Normal Distribution function.

[pic]

The gamma distribution has two parameters, a>0 and b>0. The standard gamma distribution has b=1. The Gamma Distribution function is based on the standard gamma function, and accepts an argument with a quantile value and, optionally, a value for the shape parameter a, which as a default value of 1. It returns the probability that an observation from a standard gamma distribution is less than or equal to the specified X. gammaDist is the inverse of gammaQuant.

[pic]

|The Gamma Density function (gamma pdf) accepts an |[pic] |

|argument whose value is a quantile. Optionally, you can | |

|specify a value for the shape parameter a, which has the | |

|default value 1. The figure to the right shows the shape | |

|of gamma probability density functions for various values| |

|of a. The standard gamma density function is strictly | |

|decreasing when a ≤ 1. | |

When a > 1 the density function begins at zero when X is 0, increases to a maximum and then decreases.

[pic]

The Gamma Quantile function accepts a probability argument p, and returns the pth quantile from the standard gamma distribution with the shape parameter you specify. gammaQuant is the inverse of gammaDist.

[pic]

The beta Distribution has a positive density only for an X interval of finite length, unlike normal and gamma which have positive density over an infinite interval. The beta distribution has two parameters, a>0 and b>0, and constants A ≤ X ≤ B that define the interval for which the distribution has values. The Beta Distribution function accepts the response variable argument, whose range defines A and B. You can specify values for the shape and scale parameters, or use the default values of 1.

|The standard beta distribution occurs when A = 0 and B |[pic] |

|= 1. | |

|[pic] | |

|The standard beta distribution is sometimes useful for | |

|modeling the probablistic behaviour random variables | |

|such as proportions constrained to fall in the interval| |

|0, 1. Examples of densities for several combinations of| |

|a and b are shown in the figure to the right. | |

[pic]

The Beta Quantile function accepts a probability argument p, and arguments or its shape and scale parameters. It returns the pth quantile from the standard beta distribution. betaQuant is the inverse of betaDist.

[pic]

Analyze Menu - Distribution of Y: Continuous Variables

More Moments

The More Moments popup command (Figure 1) in the Moments report adds, Sum, Variance, Skewness, Kurtosis, and CV (coefficient of variation, which is the standard deviation divided by mean and multiplied by 100) to the report.

Figure 1 Moments Table with Optional Additional Moments

[pic]

Stem and Leaf Plot

The Stem and Leaf popup command at the top of the histogram (Figure 2) does the stem and leaf plot invented by Tukey. The stem and leaf plot is a variation on the histogram. It was developed in the days when computers were rare. Each line of the plot has a Stem value that is the leading digits of a range of column values. The Leaf values are made from the next-in-line digits of the values.

To reconstruct the data values from the plot, join the stem and leaf and use the scale factor. For example, on the first line of the stem-and-leaf plot for Weight shown to the right in Figure 2, you can read the value 172 (17 from the leaf and 2 from the stem). On the last line of the plot you can read 64 and 67 as values.

The plot actively responds to clicking and brushing. When you click on leaf numbers they show in bold, and the corresponding rows in the data table and histogram bars are highlighted.

Figure 2 Stem and Leaf Plot

[pic]

Test Mean=value

The Test Mean=value popup menu command (Figure 3) has a dialog. that allows you to do a z-test if you specify the true standard deviation of the response, and to suppress or invoke the Wilcoxon signed-rank test

|Figure 3 Dialog for Test Mean=value option |t-test and no Signed-rank |

|[pic] |[pic] |

| |z test using specified sigma, |

| |sign rank checked |

| |[pic] |

|Outlier Box Plot |Figure 4 Outlier Box Plot Options |

|Figure 4 shows two new options for the Outlier Box |[pic] |

|Plot: | |

|• The Outlier Box Plot shows the means diamond. | |

|Previously, there was no mean shown in the default | |

|Distribution graphs. | |

|• The Outlier Box Plot jitters points so that when they| |

|have the same value, you can see more than one point. | |

|Capability Analysis | |

The Set Spec Limits popup menu option allows you to enter your own value for sigma, the process standard deviation.

Normal Quantile Plot

The Normal Quantile plot now shows the line of the mean and standard deviation and also the Lillifor’s bounds for normality. A probability scale appears at the top. Figures 5-8 explain normal quantile plots for four simulated distributions.

|Figure 5 |[pic] | |In the middle, the uniform distribution |

|Uniform | | |is steeper (less dense) than the normal.|

|Distribution | | |In the tails, the uniform is flatter |

| | | |(more dense) than the normal. In fact |

| | | |the tails are truncated at the end of |

| | | |the range, where the normal tails extend|

| | | |(infinitely). |

|Figure 6 |[pic] | |The normal distribution has a normal |

|Normal | | |quantile plot that tends to follow a |

|Distribution | | |straight line. The points at the end |

| | | |have the highest variance and are most |

| | | |likely not to fall near the line. This |

| | | |is reflected by the flair in the |

| | | |confidence limits near the ends. |

|Figure 7 |[pic] | |The exponential distribution is skewed, |

|Exponential | | |that is, one-sided. The top tail runs |

|Distribution | | |steeply past the normal line; it is |

| | | |spread out more than the normal. The |

| | | |bottom tail is shallow, and much denser |

| | | |than the normal. |

|Figure 8 |[pic] | |The middle portion of the double |

|Double | | |exponential is denser (more shallow) |

|Exponential | | |than the normal. In the tails, it |

|Distribution | | |spreads out more (is steeper) than the |

| | | |normal. |

[pic]

Analyze Menu - Distribution of Y: Nominal/Ordinal

Test Probabilities for Nominal Distributions

The popup command at the top of a histogram for nominal and ordinal variables has the Test Probabilities command. This command displays the dialogs shown in Figure 9 for the user to enter hypothesized probabilities. When you click Done, the Likelihood Ratio and Pearson chi-square tests are calculated for those probabilities.

The entries are scaled so that the probabilities sum to one. Thus the easiest way to test that all the probabilities are equal is to enter a 1 in each field. If you want to test a subset of the probabilities, then you leave blank the levels that are not involved, and JMP substitutes estimated probabilities.

Figure 9 Test Probabilities Dialog

[pic]

Popup options for Frequency Table

The Frequencies table popup menu (Figure 10) lets you select which columns to display in the table: Count Probability, StdErr Prob, and Cum Probability. The new item is the standard error of the probability, computed as sqrt(p*(1 – p)/n). The default table doesn’t show the standard error.

Figure 10 Frequencies Table

[pic]

[pic]

Analyze Menu - Fit Y by X: Continuous by Continuous

Nonpar Density

Nonpar Density has a new feature to do mesh (surface) plots of the density as shown in Figure 11. Also, There is also an option to turn off the 5% contours, leaving only the 10% lines.

The default kernel width has changed to .5 * stdDev * n**(1/6), which is half the old default. With a better default kernel, the slider controls to change the kernel width are now only revealed by checking Kernel Control option in the menu.

Figure 11 Mesh Plot with Nonparametric Density

[pic]

Fit Transformed

The Fitting menu now has the command Fit Transformed. This displays a dialog with choices for both the Y and X transformation. Transformations include: log, square root, square, reciprocal, and exponential.

The fitted line is plotted on the original scale as shown in Figure 12. The regression report is shown with the transformed variables, but an extra report shows measures of fit transformed in the original Y scale if there is a Y transformation.

Figure 12 Fitting Transformed Variables

|[pic] |[pic] |

|[pic] |[pic] |

Other Changes to the Fit Y by X Platform for Continuous Y and X

• A Plot Residual command is now offered for Fit line, Fit Polynomial, and Fit Transformed.

• The equation of fit is shown for Fit line, Fit Polynomial, and Fit Transformed (see the analysis report in Figure 12).

[pic]

Analyze Menu - Fit Y by X: Continuous by Nominal

Normal Quantile Plot

The normal quantile plot shows difference in means by the vertical separation of lines, and differences in variance by the differences in slope. For example, in the Taguchi.jmp data, with HTime = –1 the variance of Shrink is smaller than with HTime = 1. With Gate, the distributions seem different, leading to different box plots, but the overall variance is about the same across the Gate groups.

Figure 13 Normal Quantile and Box Plot Options for Oneway Analysis of Variance

[pic]

The Matching Variables Command

The Matching Variable command in the Analysis popup menu addresses the case when the data in a grouped analysis come from matched (paired) data. For example, observations in different groups may come from the same subject. When this happens, the statistics from the grouped F and t tests are not right. A special case of this leads to the paired t-test. However, the paired t-tests in JMP apply only when the data is organized with the pairs in different variables, not in different observations.

The Matching Variable command does two things:

• It fits an additive model that includes both the grouping variable (the X variable in the Fit Y by X analysis), and the matching variable you choose. It uses an iterative proportional fitting algorithm to do this. This algorithm makes a crucial difference if there are hundreds of subjects, because the equivalent linear model would be very slow and require huge memory resources.

• It draws lines between the points that match across the groups. If there are multiple observations with the same match id, lines are drawn from the mean of the group of observations. A Display popup menu item called Matching Lines can toggle these lines on and off, leaving just the analysis table.

For example, consider the DOGS data table after stacking the LogHist0, LogHist1, LogHist3 and LogHist5 columns into the column called time, as shown in the data table in Figure 14. To see the results, chose Fit Y by X with Y as Y and time as X. Then select the Matching Analysis command from the Analysis popup menu beneath the plot and specify id as the matching variable. The match lines connect each subject’s responses over time.

Figure 14 Matching Variables Example

|[pic] |[pic] |

The analysis in Figure 15 shows the time and id effects with F tests. These are equivalent to the tests you get using the Fit Model platform, however you would have to run two models, one with the interaction term and one without.

If there are only two levels, then the F test is equivalent to the paired t-test.

Figure 15 Matching Variables Analysis Report

[pic]

Another use for the feature is to do parallel coordinate plots when the variables are scaled similarly.

Means Diamonds Have 95% Overlap marks

Overlap marks drawn in the means diamonds are [pic] • CI above and below the group mean. For groups with equal sample sizes these marks show if the two group means are significantly different at the 95% confidence level. Figure 16 shows a Means Diamond when all Display popup menu options are in effect.

Figure 16 Information Given by Means Diamonds

[pic]

Other Changes to the Fit Y by X for Continuous Y and X

• The t-Test Report has upper and lower 95% confidence limits.

[pic]

Analyze Menu - Fit Y by X: Nominal/X by Nominal Y

Cochran-Mantel-Haenszel Test

The contingency analysis platform has the Cochran-Mantel-Haenszel test for testing if there is a relationship between 2 categorical variables after blocking across a third classification.

The Cochran-Mantel-Haenszel is a command on the popup menu at the top of the mosaic chart (Figure 17). When you select the command a dialog lets you select a grouping variable.

The example in Figure 17 uses the HOTDOGS (hotdogs.jmp) data table from the sample data. The contingency table analysis for Taste by Type shows a marginally significant chi-square probability of about .07.

Figure 17 Cochran-Mantel-Haenszel Command

[pic]

Figure 17 continued next page

Figure 17 (continued) Cochran-Mantel-Haenszel Command

[pic]

However, if you stratify on fat to protein ratio (values 1 to 4), Taste by Type show no relationship at all (Figure 18). The adjusted correlation between them is .03, and the chi-square probability associated with the with the general association of categories is .28.

Figure 18 Cochran-Mantel-Haenszel Tests Report

[pic]

[pic]

Analyze Menu - Fit Model

Standard Least Squares Parameter Estimates Details

The popup menu in the parameter estimates report (Figure 19) lets you include or suppress the standard error, t-ratio, significance p-value, 95% confidence limits, the standardized beta, and the variance inflation factor.

Figure 19 Options to Show or Suppress Columns in the Parameter Estimates Table

[pic]

[pic]

Screening: Prediction Profile

There are several new ways to control the Prediction Profiler shown in Figure 20:

• OPTION (or ALT)-click in the graph or horizontal axis area to get a dialog that lets you set the current factor value, the number of grid points, and the scale interval for that factor.

• OPTION (or ALT)-click in the left axis to change the Y response scale.

• The Confidence Intervals popup menu in the Prediction Profile lets you show or not show confidence intervals.

• The Most Desirable in Grid popup menu command in the Prediction Profile lets you search the grid for the most desirable factor setting.

• The dollar ($) popup menu lets you save a grid of predicted values in a data table.

Figure 20 Options to Control the Prediction Profiler

[pic]

Screening: Effect Screening Option for Coded Estimates

The effect screening reports (Figure 21) have major revisions to evaluate the effects in a model in a way invariant to scaling. The revisions are important in testing effects as if they had been coded to be orthogonal. In addition, when effects are not orthogonal, this option gives you a way to make them orthogonal so that they can be treated as a population of independent and identically distributed effects for normal quantiles plotting.

• The Transformed Parameter Estimates report appears open initially and shows both the original and transformed estimates.

• The report column Orthog Coded contains the orthogonalized estimates. If the design was orthogonal and balanced, then these estimates will be identical to the original estimates. If they are not, then each effect’s contribution is measured after it is made orthogonal to the effects before it.

• The old report column called Normalized Estimate is now titled Orthog t-Test. The column titled Prob>|t| is the significance level or p-value associated with the values in the Orthog t-Test column.

• The report column Orthog t-Test contains the estimates divided by their standard error. These are equivalent to Type-1 Sequential tests. The significance level (p-value) is shown beside it. This is the column that is plotted on the Normal Plot Y axis, labeled “Normalized Estimates (Orthog t).”

• The Scaled Estimate report column has been removed.

• The Pareto plot is now done with respect to the normalized estimates, which are more appropriate, especially when quadratic and crossed uncoded terms are present.

• An additional warning line appears for non-orthogonal designs stating “Each Orthog Estimate is conditioned on the effects before it.”

Figure 21 Comparison of Old and New Effect Screening Reports

|old report |new report |

|[pic] |[pic] |

old Pareto Plot

[pic]

New Pareto Plot

[pic]

Screening: Contour Profiler

The check-mark popup menu on the screening platform has the Contour Profiler option that brings up an interactive contour profiling facility. This is useful for optimizing response surfaces graphically.

Figure 22 shows an annotated example of the Contour Profiler for the TIRETREAD sample data. To see this example, run a response surface model with ABRASION, MODULUS, ELONG, and HARDNESS as response variables, and SILICA, SILANE, and SULFUR as factors. The following features are shown in Figure 22:

• There are slider controls and edit fields for both the X and Y variables.

• The Current X values generate the Current Y values. The Current X location is shown by the crosshair lines on the graph. The current Y values are shown by the small red diamonds in the slider control.

Figure 22 The Contour Profiler

[pic]

• The other lines on the graph are the contours for the responses, as set by the Y slider controls, or entering values in the Contour column. There is a separately colored contour for each response (4 in this example for the response variables).

• You can enter Low and High limits to the responses, which results in a shaded region. If you click and drag from the side zones of the Y sliders, this will set the limits, or you can enter in the Lo Limit or Up Limit columns.

• If you have more than 2 factors, then you use the check box on the right to switch the graph to other factors.

• OPTION (ALT) click on the slider control to change the scale of the slider (and the plot too if its an active X variable).

• For each contour, there is a dotted line in the direction of higher response values, so that you get a sense of direction.

|• The Update Mode determines whether |Figure 23 Optional Mesh Plots With the Contour Profiler[pic] |

|dragging the sliders produces a continuous | |

|update of the graph, or if it waits for the| |

|mouse to be released to update. On fast | |

|machines, Continuous mode is the best | |

|choice, since it calculates fast enough. | |

|Optionally, you can display Mesh Plots as | |

|in Figure 23 with the Contour Profiler. The| |

|mesh plots are displayed by default. The | |

|Surface Plot option in the popup menu in | |

|the Contour Profiler hides or displays the | |

|mesh plots. | |

[pic]

Analyze Menu - Fit Nonlinear

Two new save menu commands: Save Pred Confid and Save Indiv Confid let you save confidence intervals using the asymptotic linear approximation.

[pic]

Analyze Menu - Correlation of Y’s

There is an additional principal component command to use the covariance matrices or uncentered and unscaled in the computations instead of the correlation matrix.

[pic]

Analyze Menu - Cluster

The rows in the data table and in the dendrogram are colored and marked. The color advances in the dendrogram tree for as long as the colors agree. The Color/Marker option is now a state that forces immediate recoloring as the diamond control is moved. Also, you can use the brush tool on the dendrogram.

[pic]

Graph Menu - Control Charts

Variability Analysis (Gage R&R Charts)

Variability (Gage R&R) analysis is a new feature in the Control Charts platform.

In a Gage R&R analysis, a number of supposedly identical parts are taken from a production line. Each one is measured by a number of operators a number of times using different measuring instruments. You want to know the magnitudes of the variation due to operators, parts, and instruments. In the same way that a Shewhart control chart can identify processes which are going out of control over time, a gage R&R chart can help identify operators, instruments, or part sources that are systematically different in mean or variance.

Gage refers to gages (gauges) or instruments and R&R refers to repeatability and reproducibility. The gages or instruments that take measurements in a manufacturing process are subject to variation. Too much variation in the measurement system can mask variation in the process (part-to-part variation).

• One type of measurement variation is caused by conditions inherent in gages. This variation, know as repeatabilty, results when one person measures the same characteristic or part several times with the same gage. Repeatability is the measurement-to-measurement variation.

• Another type of measurement variation, known as reproducibility, occurs when different individuals (operators) using the same gage take measurements on the same part. Reproducibility is the operator-to-operator variability.

The Gage R&R analysis is a way to compare the part-to-part variation with operator-to-operator and measurement-to-measurement variation. The analysis uses a standard variance-components model.

In JMP, a variability analysis can have 2 or 3 factors, and there is a choice of crossed or nested model. The result is a variability (multi-var) chart, an Analysis of Variance, Variance Components report, and computation of a discrimination ratio. For 2 factor crossed or nested designs JMP also displays a Gage R&R table.

Note: This analysis requires balanced data.

You can see a variability analysis for the abrasion example by using the Date and Shift variables in the abrasn2.jmp table to represent the roles of ‘operator’ and ‘part’ in the variability analysis. To do the analysis, choose Control Charts from the Graph menu. When the control chart dialog appears, first click the Variability button. This displays the variability analysis dialog shown in Figure 24.

Complete the dialog as shown and click Chart to see a Gage R&R report for a

2-way crossed design (Figure 25).

Figure 24 Control Chart and Variability Analysis Dialogs

[pic]

Gage R&R Charts

The Variability Chart in Figure 25 shows the average abrasion measurement with max and min bars for each shift on each day. The check-mark popup menu for the Variability Chart gives options to connect the nested group means, and to show the overall group means with a dotted line.

In this example you might be concerned that difference in average measurements between shifts “A” and “B” on 4/30/95 would make it impossible to detect existing variation in the manufacturing process.

Figure 25 Variability Chart and Gage R&R Analysis

[pic]

Gage R&R Variability Report

A Gage R&R report is also given. Computations in the Gage R&R table use variance estimates, s2, given in the Variance Component Estimates table and the Sigma Multiple (u) and Tolerance you specify (see Figure 24). The Measurement Unit Analysis in the table (Figure 26) lists these quantities:

• Repeatability (EV) - equipment variation, computed as u [pic]

• Reproducibility (AV) - operator to operator variation, computed as u [pic]

• interaction (IV) - interaction of operator and part computed as u [pic]

• Gage R&R (RR) - measure of repeatability and reproducibility, computed

[pic]

• Part variation (PV) computed as u [pic]

• Total Variation (TV) - computed as [pic] )

In the Gage R&R report, % Tolerance lists the Measurement Unit Analysis for EV, AV, IV, RR, and PV divided by the tolerance you specify on the variablility chart dialog.

The Analysis of Variance performs a set of F tests appropriate for a variance components model. The denominator term is constructed to have the correct expectation to test each effect. Low p-values indicate the variance is positive.

The Discriminant Ratio characterizes the relative usefulness of a given measurement for a specific product. It compares the total variance of the measurement, M, with the variance of the measurement error, E. The Discriminant Ratio, D, is computed

D = sqrt((2*M/E) – 1)

A rule of thumb is that when the Discriminant Ratio is less than 4 the measurement cannot detect product variation, so it would be best to work on improving the measurement process. A Discrimination Ratio greater than 4 adequately detects unacceptable product variation.

Figure 26 Gage R&R Report

|[pic] |Guidelines (Barrentine, 1991) |

| |for acceptable % RR are: |

| |< 10% excellent |

| |11% - 20% adequate |

| |21% - 30% marginally acceptable |

| |> 30% unacceptable. |

|[pic] | |

|[pic] |[pic] |

Barrentine (1991), Concepts for R&R Studies, Milwaukee, WI: ASQC Quality Press.

[pic]

Index

A

Analyze menu

Cluster 11, 33

Correlation of Y’s 11, 33

Distribution of Y 8, 17-21

Fit Model 9-11, 28-32

Fit Nonlinear 11, 32

Fit Y by X 8-9, 22-28

Annotate Tool (frame and background options) 12

beta distribution functions (calculator) 16

B

binomial random number function (calculator) 14

C

calculator

efficiency 7

probability functions 7, 15

random number functions 7, 12-14

capability analysis (Distribution of Y command) 8

Cauchy random number function (calculator) 13

Close Data Windows command (Windows menu) 12

Cluster command (Analyze menu) 11, 33

Cochran-Mantel-Haenzel test (Fit Y by X command) 9, 27-28

coding convention for screening model 9-10, 31-32

coefficient of variation 8, 17

confidence limits for parameter estimates 9

contour profiler (screening model) 11, 31-32

Control Charts (Graph menu) 11, 33-36

Correlation of Y’s command (Analyze menu) 11, 33

covariance matrix for principal components analysis 11

customer support 4

CV (coefficient of variation) 8, 17

D

Data table

selected rows and columns 6

short numeric preference option 6

desirability functions (screening model) 10, 29

discrete probability tests (Distribution of Y) 8, 21

discriminant ratio (variability analysis) 36

Distribution of Y command 17-21

additional moments 8, 17

capability analysis 8

jittered points 8

Lillifor’s confidence bounds on normal quantile plot 8, 19-20

outlier box plot with jittered points 19

stem-and-leaf plot 8, 18

test discrete probabilities 8, 21

z-test 8, 18

DOE improvements 7

E

Excel files (Import command) 7

exponential random number function (calculator) 13

F

FACS file (Import command) 6

File menu

Import command for Excel files 7

Import command for FACS file 6

Open command for SAS data sets 6

Fit Model command (Analyze menu) 9-11, 28-32

fit model dialog 28

parameter estimates table 28

screening model 9-11, 29-32

standard least squares 9, 28

studentized beta 28

variance inflation factor 28

Fit Nonlinear command (Analyze menu) 11, 32

Fit Y by X command 8-9, 22-28

Cochran-Mantel-Haenzel test 9, 27-28

equation of fit added to report 8, 23

jitter display options 9

matching analysis command 9, 24-26

mesh plot for nonparametric density 8, 22

normal quantile plot shows slopes 9, 24

overlap marks on means diamonds 9, 26

plot residuals for regression 8, 23

regression on transformed numeric variables 8, 22

Frequency table (Distribution of Y) 8, 21

G

gage R&R analysis (see variablility charts)

gamma distribution functions (calculator) 16

gamma random number function (calculator) 13

geometric random number function (calculator) 14

Graph Menu (Control Charts variability analysis) 11, 33-36

gray background (preference option) 6

Group/Summary command statistical options 7

H

Hardware and Software Requirements 2

I

Import command

Excel file (Mac only) 7

FACS file 6

installing JMP 4

J

jitter option

Distribution of Y 8

Fit Y by X command 9

K

kurtosis (in moments table) 8, 17

L

Lillifor’s confidence bounds 8, 19-20

M

matching model analysis (Fit Y by X command) 9, 24-26

means diamonds (overlap marks) 9, 26

median (Group/Summary platform) 7

memory recommended 2-3

mesh plot

Fit Y by X nonparametric density 8, 22

screening model 32

mesh plots (Fit Model platform screening model) 11

N

negative binomial random number function (calculator) 14

normal distribution functions (calculator) 15

normal quantile plot (Fit Y by X command) 9, 24

normal random number function (calculator) 13

normalized estimates in screening model 10, 30

O

Open command for SAS data sets 6

optimizing response surfaces (screening model) 11, 29-32

orthogonal t-test in screening model 10, 30

orthogonalized estimates in screening model 10, 30

P

parameter estimates table (standard least squares) 28

Pareto plots in screening model 10, 30

percent total (Group/Summary platform) 7

plot residual option (Fit Y by X regression) 8, 23

Poisson random number function (calculator) 13

prediction profiler (screening model) 10, 29

Preferences option

gray background 6

short numeric choice enabled 6

press statistic 9

principal components analysis (Correlation of Y’s command) 11

probability functions (calculator) 7, 15

R

random number functions (calculator) 7, 12-14

range (Group/Summary platform) 7

regression equation 8, 23

S

SAS data sets, use Open command 6

save commands for nonlinear fit 11, 32

screening model (Fit Model platform)

coding conventions 9-10, 29-30

contour profiler 11, 31-32

desirability functions 10, 29

mesh plots 11, 32

normalized parameter estimates 10, 30

optimizing response surfaces 11, 29-32

orthogonal t-test 10, 30

orthogonalized parameter estimates 10, 30

Pareto plots 10, 30

prediction profiler 10, 29

saving grid of predicted values 11

short numerics (new preference option) 6

shuffle random number function (calculator) 13

skewness (in moments table) 8, 17

standard least squares (Fit Model platform)

confidence limits for parameter estimates 9

parameter estimates table 9, 28

press statistic 9

standardized regression coefficients 9, 28

variance inflation factor 9

standardized regression coefficients 9, 28

stem-and-leaf plot 8, 18

Summary Platform statistical options 7

System Hardware and Software Requirements 2

T

technical support 4

Tools menu (Annotate Tool) 12

training and education 4

transformed numeric variables for regression 8, 22

triangular random number function (calculator) 13

V

variability analysis (Control Charts) 33-36

analysis of variance 36

control chart dialog 34

discriminant ratio 36

Gage R&R chart and table 35

variance components estimates 36

variability charts (Control charts command) 11

variance inflation factor 9, 28

W

Windows menu (Close Data Windows command) 12

World Wide Web for information and technical support 5

Z

z-test (Distribution of Y) 8, 18

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download