Rev 2/26/14

Stat-Ease
Handbook
for
Experimenters
A concise collection of handy tips
to help you set up and analyze
your designed experiments.

Version 09.01.03

Rev 2/26/14

Acknowledgements
The Handbook for Experimenters was compiled by the statistical staff at
Stat-Ease. We thank the countless professionals who pointed out ways to
make our design of experiments (DOE) workshop materials better. This
handbook is provided for them and all others who might find it useful to
design a better experiment. With the help of our readers, we intend to
continually improve this handbook.

NEED HELP WITH A DOE?
For statistical support
Telephone: (612) 378-9449
E-mail: stathelp@statease.com

Copyright 2014 Stat-Ease, Inc.
2021 East Hennepin Ave, Suite 480
Minneapolis, MN 55413
www.statease.com

yet they allow you to identify interactions in your process. Section 1 is meant to be used BEFORE doing your experiment. Traditionally. Service industries have also benefited by obtaining data from their process and analyzing it appropriately. experimentation has been done in a haphazard one-factor-ata-time (OFAT) manner. electronics.Rev 2/26/14 Introduction to Our Handbook for Experimenters Design of experiments is a method by which you make purposeful changes to input factors of your process in order to observe the effects on the output. automotive. require only a minimal number of runs. It provides guidelines for design selection and evaluation. thus improving quality. On the other hand. pharmaceutical. factorial designs are a very basic type of DOE. This method is inefficient and very often yields misleading results. DOE’s can and have been performed in virtually every industry on the planet—agriculture. Section 3 contains various statistical tables that are generally used for manual calculations. Section 2 is a collection of guides to help you analyze your experimental data. hard goods manufacturing. etc. -The Stat-Ease StatHelp Team . This information leads you to breakthroughs in process understanding. even non-users can find a great deal of valuable detail on DOE. reducing costs and increasing profits! We designed this Handbook for Experimenters to use in conjunction with our Design-Ease® or Design-Expert® software. chemical. However.

................................................. 2-7 Statistical Details on Diagnostic Measures ................................................... 1-19 Matrix Measures ..................1%) ............................... 3-4 Normal Distribution Two-Sided Tolerance Limits (K2) .....................................................................................................5%..................................................... 1-7 Factorial Design Worksheet ............ 1-11 Response Surface Design Worksheet ... 1-16 Mixture Design Selection .............................. 3-18 Distribution-Free One-Sided Tolerance Limit .......................... 2-11 Optimization Guide ............................. 2-14 Inverses and Derivatives of Transformations ................................................................................................................................................................................... 3-1 T-table (One-Tailed/Two-Tailed) ....... 2-3 Combined Mixture / Process Analysis Guide ....................................................................................... 2.......................................................................... 3-2 Chi-Square Cumulative Distribution ............................. 1-13 Number of Points for Various RSM Designs ............Rev 2/26/14 Table of Contents DOE Process Flowchart Cause-and-effect diagram (to fish for factors) Section 1: Designing Your Experiment DOE Checklist ............................................ 3-20 .............................................. 2-15 Section 3: Appendix Z-table (Normal Distribution) ......................................... 1-8 Factorial Design Selection .... 1-3 Impact of Split-Plot Designs on Power .............. 3-3 F-tables (10%.............................................................. 1-5 Binomial Data ......... 1%...... 0............................ 1-22 Section 2: Analyzing the Results Factorial Analysis Guide . 1-12 RSM Design Selection.................................... 3-14 Distribution-Free Two-Sided Tolerance Limit ........ 1-2 Power Requirements for Two-Level Factorials .......... 1-9 Details on Split-Plot Designs ..............5%........................................................................................ 1-6 Standard Deviations... 5%........................................................ 1-18 Design Evaluation Guide ... 1-17 Combined Design Selection .............................. 3-10 Normal Distribution One-Sided Tolerance Limits (K1) ......................................................................... 2-1 Response Surface / Mixture Analysis Guide ....................... 0........... 2-6 Residual Analysis and Diagnostic Plots Guide ............................................................... 1-1 Factorial DOE Planning Process .......................... 1-20 Fraction of Design Space Guide ... 1-15 Mixture Design Worksheet .........

consider a split plot. .Rev 2/26/14 DOE Process Flowchart Define Objective and Measurable Responses Begin DOE Checklist (p1-1) Brainstorm Variables Next page illustrates how to fish for these Mix Inputs Mixture Worksheet (p1-16) & Design Selection (p1-17) Analysis Guide (p2-3) +Process Combined Design (p1-18) Analysis Guide (p2-6) Stage Screen/ Characterize Factorial Worksheet (p1-8) & Design Selection (p1-9) Analysis Guide (p2-1) Optimize RSM Worksheet (p1-12) & Design Selection (p1-13) Analysis Guide (p2-3) Residual Analysis and Diagnostic Plots Guide (p2-7) Optimization Guide (p2-13): Find Desirable Set-up that Hits the Sweet Spot! If some factors are hard-to-change (HTC).

as many as a dozen. o Idea for prioritizing variables: Give each evaluator an imaginary 100 units of currency to ‘invest’ in their favorites.  Note factors that are hard to change. pare the group down to three or so key people who can then critically evaluate the collection of variables and chose ones that would be most fruitful to experiment on. Machinery.Alex Osborne .  Variables not chosen should be held fixed if possible.  Alternative: To be more participative. Choosing variables to experiment on and what to do with the others:  For the sake of efficiency. start by asking everyone to note variables on sticky notes that can then be posted on the fishbone diagram.” . and o Assign one to be leader. Consider either blocking them out or including them for effects assessment via a split plot design. Method. for example.Rev 2/26/14 Cause-and-effect diagram (to fish for factors) Response (Effect): ______________ ______________ ______________ Suggestions for being creative on identifying potential variables:  Label the five big fishbones by major causes. o Another individual should record all the ideas as they are presented. Material. People and Environment (spine).  Gather a group of subject matter experts. who will be responsible for maintaining a rapid flow of ideas.  Keep a log of other variables that cannot be fixed but can be monitored. “It is easier to tone down a wild idea than to think up a new one. Tally up the totals from top to bottom.

Rev 2/26/14 Section 1: Designing Your Experiment .

Rev 2/26/14 This page left blank intentionally as a spacer. .

such as material. etc. Make plans for follow-up studies. Negotiate time. o Blocks (to filter out known source of variation. o Estimate difference Δ (delta) generated in response(s) o Be bold. Select design (see details in Handbook). Identify uncontrolled variables: Can they be monitored? Establish procedures for running an experiment. o Determine the power (or size by fraction of design space— FDS—if an RSM and/or mixture). Choose a model based on subject matter knowledge of relationship between factors and responses. Go over details of the physical setup and design execution. Identify response variables and how they will be measured.Rev 2/26/14 DOE Checklist                Define objective of the experiment. Choose low and high level of each factor (or component if a mixture). Determine how to hold non-DOE variables constant. o Center points (or centroid if a mixture). but avoid regions that might be bad or unsafe. Decide which variables to investigate (brainstorm—see fishbone in the Handbook preface). o Invest no more than one-quarter of your experimental budget (time and money) in the first design. Specify: o Replicates. Take a sequential approach.). Perform confirmation tests. Be flexible! Discuss any other special considerations for this experiment. equipment. day-to-day differences. 1-1 . material and budgetary constraints. Evaluate design (see details in Handbook): o Check aliasing among effects of primary interest.

Identify opportunity and define objective. • Evaluate power (ideally greater than 80%). a. 1. you will determine how many runs to budget. Estimate experimental error (σ) for each response. 4. This information is needed for EACH response. Define the change (Δy) that is important to detect for each response. • Examine the design to ensure all the factor combinations are reasonable and safe (no disasters!) 1-2 . • If fractionated and/or blocked. Based on a projected signal-to-noise ratio. This is your “signal. This is your “noise. Use the signal to noise ratio (Δy/σ) to estimate power. evaluate aliases with the order set to a two-factor interaction (2FI) model. State objective in terms of measurable responses. • Are any factors hard-to-change (HTC)? If so consider a split-plot design. 2. If the design is a split-plot.) The factor ranges must be large enough to (at a minimum) generate the hoped-for change(s) in the response(s).” b. 3.” c. Select a factorial design (see Help System for details). See the next page for an example on how to calculate signal to noise. consider the trade-off in power versus running a completely randomized experiment.Rev 2/26/14 Factorial DOE Planning Process This four-step process guides you to an appropriate factorial DOE. Select the input factors to study. (Remember that the factor levels chosen determine the size of Δy.

1-3 . General Procedure: 1. This is the change in the response that you want to detect.* Continue this process until you achieve the desired power. then you must find a way to decrease noise (σ).8. historical data or experience (just make a guess!). background (white or cyan) and typeface (Arial or Times New Roman). Set up your design and evaluate its power based on the signal-tonoise ratio (Δ/σ). Determine the signal delta (Δ). *(If it’s a fraction. A 23 design (8 runs) is set up to minimize time needed to read a 30-word paragraph. then chose a less-fractional design for a better way to increase runs—adding more power and resolution. What’s the threshold value that arouses interest? That’s the minimum signal you need to detect. Just estimate it the best you can—try something! 2. • The signal to noise ratio (Δ/σ) is 1/. We want the power to detect this to be at least 80%. This is the noise: σ = 0. or both. • A prior DOE reveals a standard deviation of 0. This is the signal: Δ = 1. Following the procedure above.8 = 1. increase the signal (Δ). Bounce numbers off your management and/or clients. consider adding more runs or even replicating the entire design.) Example: What is the ideal color/typeface combination to maximize readability of video display terminals? The factors are foreground (black or yellow). If the minimum runs exceeds what you can afford. Estimate the standard deviation sigma (σ)—the noise—from: • • • • repeatability studies control charts (R-bar divided by d2) analysis of variance (ANOVA) from a DOE. 3.25.Rev 2/26/14 Power Requirements for Two-Level Factorials (p1/5) Purpose: Determine how many runs you need to achieve at least an 80% chance (power) of revealing an active effect (signal) of size delta (Δ).8 seconds in readings. starting with a ridiculously low improvement in the response and working up from there. determine the signal-to-noise ratio: • A 1-second improvement is the smallest value that arouses interest from the client. If it’s less than 80%.

25 signal/noise ratio – not good enough.  3.25: Mission accomplished!  1-4 . Add a 3rd replicate (blocked) to the design (for a total of 24 runs) and evaluate. which falls far short of the desired 80%. Power is now over 80% for the ratio of 1.Rev 2/26/14 Use Design-Expert to Size a Regular Two-Level Design for Adequate Power: 1.5% for the 1. For the 8-run regular 23 design enter the delta and sigma. 2. The probability of detecting a 1 second difference at the 5% alpha threshold level for significance (95% confidence) is only 27.6%. Go back and add a 2nd replicate (blocked) to the design (for a total of 16 runs) and re-evaluate the power. The program then computes the signal-to-noise (Δ/σ) ratio of 1.25. The power increases to 62.

thus by going back to Options.  However. subject matter knowledge for this example indicates that the HTC factors vary far less—by a 1-to-4 ratio—than the ETC.9% to 98. the Variance ratio of whole-plot to sub-plot variance is changed to 0. To accommodate these HTC factors in a reasonable number of runs.4%) due to being in the “subplot” part of the split-plot design. the HTC factors a.”  Fortunately.25 (=1/4). This restores adequate power —over 80%—to the HTC factors. b and c (lower-case) lose power by being restricted in their randomization to “whole plots. Three of the five suspect factors are hard to change (HTC).Rev 2/26/14 Impact of Split-Plot (vs Randomized) Design on Power: Illustration: Engineers need to determine the cause of drive gears becoming ‘dished’ (a geometric distortion).  1-5 . they select a half-fraction Split-Plot Regular Two-Level design and assess the power for a signal of 5 and a noise of 2 as shown in the screen shot below. The program then produces these power calculations: The easy-to-change (ETC) factors D and E (capitalized) increase in power (from 88.

Two-Level Factorial” worksheet. The output will show you the number of parts per run. change to detect. the alpha and power. fill in the current proportion. 3. Open the spreadsheet. 2.05 (fraction) and Δ = 0. and finally the number of runs in the design. Contact Stat-Ease via stathelp@statease. Then follow these steps.Rev 2/26/14 Special Procedure for Handling Binomial Data: If the data is binomial data. For example.xlsx”.02 (signal). On the “Binomial . Determine your current proportion and the change (signal) you want to detect (Δ). such as a measurement of pass or fail from a quality inspection or the response to a direct mailing. 1-6 .com to obtain the handy spreadsheet calculator “EDME Calculations. the normal power calculations for two-level factorials (those included in Design-Ease or Design-Expert software) do not work. then: p = 0. where p = (#of fails or passes) / (#total units). if 5% of your parts are now running defective and you want to improve this by at least 2% (to 3% or less). 1. Convert the measurement to a proportion (“p”).

the required Power (usually 80%). the Current standard deviation (estimate). 2. 4.Rev 2/26/14 Special Procedure for Handling Standard Deviation In many situations you will produce a number of parts or samples per run in your experiment. Then we recommend you compute the standard deviation of each response so you can find robust operating conditions by minimizing variability. Input your sub-group size. or optionally (if macros can be enabled).com to obtain the handy spreadsheet calculator “EDME Calculations.” which assumes that you entered the minimum amount of improvement desired.xlsx”. As shown in the screenshot below. Open the spreadsheet to the “Std Dev Two-Level Factorial” worksheet. enter the Alpha level (typically 5%). the Size of change in standard deviation you hope to achieve and the Number of runs. Make one of two choices for your objective: • “Limit the growth. Contact Stat-Ease via stathelp@statease. 1-7 . Then follow these steps.” which assumes the change you entered is how much extra variance you can tolerate. 3. 1. • “Reduce Standard Deviation. use the calculate button to calculate the sample size to reach the specified power. The standard power calculations for two-level factorials do not work in this case.

maximize. Select the input factors and ranges to vary within the experiment: Remember that the factor levels chosen determine the size of Δy. Name Units Type HTC*? Low (−1) High (+1) A: B: C: D: E: F: G: H: J: K: *Hard-to-change (versus easy-to-change—ETC) Choose a design: Type:____________________________________ Replicates: ____. • Estimate the experimental error (σ . Name Units σ Δy Δy/ σ Power Goal* R1: R2: R3: R4: *Goal: minimize.Rev 2/26/14 Factorial Design Worksheet Identify opportunity and define objective: ________________________ ______________________________________________________ ______________________________________________________ State objective in terms of measurable responses: • Define the change (Δy . 1-8 Center points: ____ .signal) you want to detect.noise) • Use Δy/σ (signal to noise) to check for adequate power. Blocks: _____. etc. target=x.

but these are aliased with each other [2FI – 2FI]. 1-9 . Randomized designs. compare the screenshots below.  Red: Resolution III designs (ME’s aliased with 2FI’s. Caution: If any responses go missing. Caution: Do not use these designs to screen for significant effects. re-sorted somewhat for usefulness. This boils downs to a go/no-go acceptance test.) Good for ruggedness testing where you hope your system will not be sensitive to the factors. Min-Run Characterize (Resolution V): Balanced (equireplicated) two-level designs containing the minimum runs to estimate all ME’s and 2FI’s. Check the power of these designs to make sure they can estimate the size effect you need.) Good for estimating ME’s and 2FI’s. Provides information on all effects.  Green: Resolution V designs or better (main effects (ME’s) aliased with four factor interactions (4FI) or higher and two-factor interactions (2FI’s) aliased with three-factor interactions (3FI) or higher. notably split-plots to accommodate hard-to-change factors by restricting their randomization into groups. Careful: If you block. All possible combinations of factor levels are run. feature more-descriptive names.Rev 2/26/14 Factorial Design Selection (p1/3) Version 9 of Stat-Ease software introduced a number of new options.  White: Full factorials (no aliases). For a quick overview of the changes. some 2FI’s may be lost!  Yellow: Resolution IV designs (ME’s clear of 2FI’s.) Useful for screening designs where you want to determine main effects and the existence of interactions. These designs are color-coded in StatEase software to help you identify their nature at a glance. then the design degrades to Resolution IV. Design-Expert V8 Software – Old Design-Expert V9 Software – New! New Design New Designs Regular Two-Level: Selection of full and fractional factorial designs where each factor is run at 2 levels.

delete columns to estimate certain interactions. To avoid this sensitivity. Plackett-Burman: OK for ruggedness testing but not much else due to extremely complex Resolution III alias structures. Via Taguchi’s ‘linear graphs’. Caution: even one missing run or response degrades the aliasing to Resolution III. Definitive Screen: A three-level design that estimates main effects. D-optimal factorial designs are recommended.all columns used for ME’s. *(Not powers of two. 1-10 . Use only with great care! Taguchi OA (“Orthogonal Array”): Saturated Resolution III arrays .Rev 2/26/14 Irregular Res V*: These special fractional Resolution V designs may be a good alternative to the standard full or Res V two-level factorial designs. Not recommended if you suspect there may be several interactions present in the system.) Optimal (Custom): Choose any number of levels for each categoric factor. Quadratic (squared) terms are aliased with sets of 2FI’s.) Min-Run Screen (Resolution IV): Estimates main effects only (the 2FI’s remain aliased with each other).) Simple Sample: Use this design choice as a tool for entering raw data to generate basic statistics (mean. Check the power. standard deviations and intervals) for a process where no inputs are intentionally varied. Multilevel Categoric: A general factorial design good for categoric factors with any number of levels: Provides all possible combinations. for example 12 runs for 4 factors. (Design also available in Split-Plot. (Design also available in Split-Plot. We recommend you not use these designs. accept the Stat-Ease software design default adding two extra runs (Min Run +2). There are no factors to enter—only a specified number of observations (runs containing one or more measured responses). If too many runs. The number of runs chosen will depend on the model you specify (2FI by default). use Optimal design.

within which the easy-to-change (ETC) factors will be randomized in subplots. Optimal (Custom): Change factors to Hard or Easy.Rev 2/26/14 Details on Split-Plot Design: Regular Two-Level: Select the number of total factors and how many of these will be hard to change (HTC). if you can afford the runs. Note the defaults for required model points and groups and. From one group to the next.) Multilevel Categoric: Change factors to Hard or Easy as shown. be sure to reset each factor level even if by chance it does not change. In that case go Back and add more Groups. the additional groups and points. 1-11 . then increase Replicates. Caution: Watch out for low power on the HTC factor(s).* The HTCs will be grouped in whole plots. The program may then change the number of runs to provide power. *(Caution: You may be warned on the power screen that “Whole-plot terms cannot be tested…” Proceed then with caution—accepting there being no test on HTC(s)—or go back and increase the runs. If you see the “Cannot test…” warning upon Continue.

Rev 2/26/14 Response Surface Method (RSM) Design Worksheet Identify opportunity and define objective: __________________ ________________________________________________ ________________________________________________ ________________________________________________ State • • • objective in terms of measurable responses: Define the precision (d . Blocks: _____. 1-12 Center points: ____ .signal) required for each response. Use d/σ (signal to noise) to check for adequate precision using FDS. Name Units d σ FDS Goal R1: R2: R3: R4: Select the input factors and ranges to vary within the experiment: Name Units Type Low High A: B: C: D: E: F: G: H: Quantify any MultiLinear Constraints (MLC’s): ________________________________________________ ________________________________________________ Choose a design: Type:____________________________________ Replicates: ____.noise) for each response. Estimate the experimental error (σ .

rotatable. Region of operability must be greater (−α. +α) Good design properties. and it’s very sensitive to outliers and missing data.  Small (Draper-Lin) Minimal design – not recommended: statistical evaluation shows that properties are marginal. etc.)  Practical alpha (α = 4th-root of k – the number of factors): Recommended for six or more factors to reduce collinearity in CCD. custom polynomial. −α) or o Min-run Res V. Optimal: Useful for building custom designs. Use when region of interest and region of operability nearly the same. Can handle categoric factors.minimizes the joint confidence interval for the model coefficients. little collinearity. (0. insensitive to (+1. orthogonal blocks.Rev 2/26/14 RSM Design Selection (p1/2) Central Composite Designs (CCD):  Standard (axial levels (α) for “star points” are set for rotatability): (0.−1) o Standard Resolution V fractional design. change factorial core of CCD to: (+1. constrained design spaces. Box-Behnken (BBD): Each factor has only three levels. 0) runs. Each factor has five levels. rotatable or nearly rotatable. so default for factorial designs) o A . 0) than region of interest to accommodate axial (0. (Not recommended for six or more factors due to high collinearity in squared terms. 0) (+α. insensitive to outliers and missing data. For 5 or more factors. cuboidal rather than rotatable.default reduces the average prediction variance.  Face-centered (FCD) (α = 1. Good design properties.minimizes the average confidence interval for the model coefficients. insensitive to outliers and missing data. Uses two exchange algorithms to find the optimal set of points to estimate the designed for model. some have orthogonal blocks. Types of optimal: o I . Does not predict well at the corners of the design space.0): Each factor conveniently has only three levels. 1-13 .−1) (−1.+1) outliers and missing data. (Best for finding effects. little collinearity. Good design properties for designs up to 5 factors: little collinearity. (Best predictions) o D . Use when region of interest and region of operability are nearly the same.

90 1.00 -0. Good design properties. Created by using a CCD for all factors except the last. insensitive to outliers and missing data.00 0.80 D :D 0.chooses the best design from Point or Coordinate exchange. o Point exchange – based on geometric candidate set of fixed coordinates o Coordinate exchange – candidate-set free: Points freely located anywhere within the experimental region. Historical: Allows for import of existing data. 0) and 4 levels of one factor versus 5 of the other. Caution: very sensitive to outliers and missing data – provides only minimum number of runs required. Region of operability must be greater than region of interest to accommodate axial runs.90 -1. Each factor has four or five levels. Hybrid: Minimal design – not recommended but good alternative to Small CCD.Number of Design Points for Various RSM Designs). 1-14 .80 A:A Pentagonal: For two factors only. this minimal-point design provides an interesting geometry with one apex (1.80 -0. The resulting models may fit past results adequately but remain useless for prediction. It may be of interest with one categoric factor at two levels to form a three-dimensional region with pentagonal faces on the two numeric (RSM) factors. o User-Defined: Generates points based on geometry of design space. this design is a good alternative to the pentagon with 5 levels of one factor versus 3 of the other. cuboidal rather than rotatable.90 0. consider BBD or FCD. maybe three – beyond that the number of design points become excessive relative to the number of coefficients needed to fit a quadratic polynomial model for response surface purposes (see table on next page . To reduce runs for more than three factors.not recommended: chooses points as far away from each other as possible. 1. Design evaluation shows little collinearity and the design is rotatable or nearly so.Rev 2/26/14 Distance based .80 -1. Miscellaneous RSM designs include the following: Three-Level Factorial: Usable for two factors. Exchange Algorithms: o Best (default) .90 0. Hexagonal: For two factors only. Be sure to evaluate this happenstance design before doing the analysis. Do not be surprised to see extraordinarily high variance inflation factors (VIF’s) due to multicollinearity. thus achieving maximum spread. which gets chosen D-optimally (becoming oddly spaced as shown in the figure).

Rev 2/26/14 Number of Points for Various RSM Designs Number Factors CCD full CCD fractional CCD MR-5 BoxBehnken Small CCD Quadratic Coefficients 2 13 NA NA NA NA 6 3 20 NA NA 17 15 10 4 30 NA NA 29 21 15 5 50 32 NA 46 26 21 6 86 52 40 54 33 28 7 152 88 50 62 41 36 8 272 154 90 60 120 51 45 9 540 284 156 70 130 61 55 10 X 286 158 82 170 71 66 20 X 562 258 348 NA 231 30 X NA 532 NA NA 496 X = Excessive runs NA = Not Available 1-15 .

signal) required for each response. Name Units d σ FDS Goal R1: R2: R3: R4: Select the components and ranges to vary within the experiment: Name Units Type Low High A: B: C: D: E: F: G: Mix Total: Quantify any MultiLinear Constraints (MLC’s): ________________________________________________ ________________________________________________ Choose a design: Type:___________________________________ Replicates: ____.Rev 2/26/14 Mixture Design Worksheet Identify opportunity and define objective: __________________ ________________________________________________ ________________________________________________ State objective in terms of measurable responses: • Define the precision (d . • Estimate the experimental error (σ .noise) for each response. • Use d/σ (signal to noise) to check for adequate precision using FDS. 1-16 Centroids: ____ . Blocks: _____.

Design is then constructed of m+1 equally spaced values from 0 to 1 (coded levels of individual mixture component).. .quadratic or 3 .. 0) o and so on to the overall centroid: (1/q.linear... . The resulting number of blends depends on both the number of components (“q”) and the degree of the polynomial “m”.) Screening designs: Essential for six or more components.  Lattice: Specify degree “m” of polynomial (1 .cubic). possibly with constraints.  Distance-Based: (See RSM design selection for details.. 0. 0) o Binary (two-part) blends: (1/2. 1/3. but not sufficient. 1/3.  Simplex screening  Extreme vertices screening (for non-simplex) 1-17 ..) Use when component ranges are not the same. ...  Centroid: Centroid always included in the design comprised of 2q-1 distinct mixtures generated from permutations of: o Pure components: (1. 0. .. 0.. 1/q. to form a simplex geometry for the experimental region). Centroid not necessarily part of design. or you have a complex region.. 1/q) Simplex Lattice versus Simplex Centroid Extreme vertices designs (in order of preference):  Optimal: (See RSM design selection for details. 2 .Rev 2/26/14 Mixture Design Selection Simplex designs: Applicable if all components range from 0 to 100 percent (no constraints) or they have same range (necessary. Creates design for linear equation only to find the components with strong linear effects. 0) o Tertiary (three-part) blends: (1/3. 1/2.)  User-Defined: (See RSM design selection for details..

For example. a second-order polynomial.Rev 2/26/14 Combined Design Selection These designs combine either two sets of mixture components. or mixture components with numerical and/or categoric process factors.* Design-Expert software will add by default: • Lack of fit points (check blends) via distance-based criteria • Replicates on the basis of leverage. At least two types of variables. the second-order two-factor interaction (2FI) model will be selected for the categoric factors.000 for the filled cupcakes!). *The model for categoric factors takes the same order as for the numeric (process). must be specified to achieve the status of a combined design. salt and chocolate  Numeric factors – the baking process: 2 for time and temperature  Categoric factors – the oven: 2 types – Easy-Bake or gas. one of them being “Mixture 1”. You must specify models for the mixtures and process factors (quadratic by default). identify the number of:  Mixture 1 components – the cake: 4 for flour. sugar and eggs  Mixture 2 components – the filling: 3 for cream cheese. For example. The optimal design option pares down the runs to the bare minimum needed to fit the combined models. if you want to mix your filled cupcake and bake it too using two ovens. water. The User Defined design option generates a very large candidate set (over 25. Therefore. 1-18 . by default the process will be quadratic.

Select the polynomial model you want to evaluate. but they should all be equal to each other.examine this. If the model is aliased.) 5. the program calculates the alias structure -. (Specifically the standard error of a model coefficient is increased by a factor equal to the square root of the VIF.Rev 2/26/14 Design Evaluation Guide (p1/3) 1. They should be the same within type of coefficient. Look at the standard errors (based on s = 1) of the coefficients. that is. For more details. First look for aliases. the standard errors associated with all the linear (first order) coefficients should be equal. For RSM and mixture designs: look at fraction of design space (FDS) graph to evaluate precision rather than power. (See FDS Guide. or the wrong set of design points was chosen.  VIFs exceeding 10 indicate the associated regression coefficients are poorly estimated due to multicollinearity.) 1-19 . specify main effects model only. Examine the table of degrees of freedom (df) for the model. Degrees of freedom for residual error must be available to calculate power. 3. 4.)  VIF of 1 is ideal because then the coefficient is orthogonal to the remaining model terms. Examine the variance inflation factors (VIF) of the coefficients: VIF = 1 1. see Power Calculation Guide. its standard error is 4 times as large as it would be in an orthogonal design. and so on. No aliases should be found. 2. when compared to the standard error for the same model coefficient in an orthogonal design. so for unreplicated factorial designs. An aliased model implies there are either not enough unique design points. You want: a) Minimum 3 lack-of-fit df. The standard errors for the cross products (second order terms) may be different from those for the linear standard errors. (If a coefficient has a VIF of 16. the correlation coefficient (Ri2) is 0.Ri2  VIF measures how much the lack of orthogonality in the design inflates the variance of that model coefficient. For example. b) Minimum 4 df for pure error. For factorial designs: Look at the power calculations to determine if the design is likely to detect the effects of interest.

Matrix Measures (for more thorough evaluation by statistical researchers) 1. Ideally the design produces a flat error profile centered in the middle of your design space. Maximum. The shape of this plot depends only on the design points and the polynomial being fit.. If possible. Condition Number of Coefficient Matrix (ratio of max to min eigenvalues. orthogonal • • κ < 100 κ < 1000 multicollinearity not serious moderate to strong multicollinearity • κ > 1000 severe multicollinearity {Note: Since mixture designs can never be orthogonal. Average. 7. Consider replicating points where leverage is more than 2 times the average and/or points having leverage approaching 1. They are the variance multipliers for the prediction interval around the mean. Repeat the “design evaluation – design modification” cycle until satisfied with the results. Then go ahead and run the experiment. Contour (or 3D Surface) Do a plot of the standard error (based on s = 1). Note: Lack-offit and replicates tend to reduce the G efficiency of a design. These are estimated by the Fraction of Design Space sample. G Efficiency – this is a simple measure of average prediction variance as a percentage of the maximum prediction variance. the matrix condition number can’t be evaluated on an absolute scale. Examine the leverages of the design points. or roots.} b. 1-20 . Go to Graphs. and Minimum mean prediction variance of the design points. of the X'X matrix): κ = λmax/λmin • κ=1 no multicollinearity.Rev 2/26/14 6. try to get a G efficiency of at least 50%. c. Average leverage = p N Where “p” is the number of model terms including the intercept (and any block coefficients) and “N” is the number of experiments. For an RSM design this should appear as either a circle or a square of uniform precision.e. i. Evaluate measures of your design matrix: a.

In an orthogonal design all correlations with other coefficients are zero. look at all the measures presented during design evaluation. To get a more balanced assessment. Scaled D-optimality .Rev 2/26/14 d. It is usually not possible to minimize all three simultaneously. primarily for algorithmic point selection. Scaling allows comparison of designs with different number of runs. e. The Doptimality criterion minimizes the variance associated with the coefficients in the model. • The trace (related to A-optimal) represents the average variance of the model coefficients. 4. b.this matrix-based measure assesses a design’s support of a model in terms of prediction capability. It is a single-minded criterion which often does not give a true measure of design quality. • The determinant (related to D-optimal) measures the volume of the joint confidence interval of the model coefficients. • The I-score (related to I-optimal) measures the integral of the prediction variance across the design space. Add additional runs manually or via the design tools in Stat-Ease software for augmenting any existing set of runs. Examine the correlation matrix of the model coefficients (derived from (X'X)-1). Examine the correlation matrix of the independent factors (comes directly from the X matrix itself). Mixture designs can never be orthogonal. 2. Modify your design based on knowledge gained from the evaluation: a. 1-21 . The smaller the scaled D-optimal criterion the smaller the volume of the joint confidence interval of the model coefficients. How close is your design to this ideal? {Note: Due to the constraint that the components sum to a constant. mixture designs can never be orthogonal. In an orthogonal design none of the factors are correlated. Choose a different design.} 3. When scaled the formula becomes: N((determinant of (X'X)-1)1/p) Where N is the number of experiments and p is the number of model terms including the intercept and any block coefficients. trace and I-score are relative measures (the smaller the better!) used to compare designs having the same number of runs. The determinant.

Note: More runs are required to get similar precision with “Pred” than “Mean”. • • • • The FDS graph can be produced for four different types of error: mean. Supply the “signal” and the “noise” and the graph will show the amount of the design region that can estimate with that precision. prediction. There is also the option to create the FDS by using either One-Sided or Two-Sided intervals. optimization is based on average trends. 1-22 . FDS is determined by four parameters: the polynomial used to model the response. and when using Diff it’s the minimum change in the response that is important to detect.Rev 2/26/14 Fraction of Design Space (FDS) Guide FDS evaluation helps experimenters size constrained response surface (RSM) and mixture designs. and “d”. The meaning of “d” changes relative to the error type selected. Smaller changes are more difficult to detect. or tolerance. Diff – recommended when searching for any change in the response. For Mean it’s the half-width of the confidence interval. Pred – best when the goal is to verify individual outcomes. Tolerance –useful for setting specifications based on the experiment. for Pred it’s the half-width of the prediction interval. An FDS greater than 80 percent is generally acceptable to ensure that the majority of the design space is precise enough for your purpose. “s” or estimated standard deviation. Mean – used when the goal of the experiment is to optimize responses using the model calculated during analysis. such as for verification DOE’s. for which the normal power calculations lose relevance. difference. for Tolerance it’s the half-width of the tolerance interval. “a” or alpha significance level.

Rev 2/26/14 Section 2: Analyzing the Results .

.Rev 2/26/14 This page left blank intentionally as a spacer.

Half-Normal Plot Pareto Chart A 9.34 90 t-Value of |Effect| Half-Normal % Probability AD A D 95 AC AD 80 70 C D 50 C 4. on the Pareto Chart pick effects from left to right. largest to smallest. terms having (Prob > F) > 0. 2-1 .e.81 16.90 Bonferroni Limit 3. Refer to the Residual Analysis and Diagnostic Plots Guide. Be sure to maintain hierarchy. Alternatively.10: not significant .) 3. c) Check for “Adeq Precision” > 4.10: not significant .41 10..05: significant . This is a signal to noise ratio (see formula in Response Surface Analysis Guide). Click the biggest effect (point furthest to the right) and continue right-to-left until the line runs through points nearest zero.00 1 0. (Not available for split plots.  Curvature and Lack of fit (if reported) should be insignificant: • P-value < 0. i. Choose ANOVA* (Analysis of Variance) and check the selected model: *(For split plots via REML—Restricted Maximum Likelihood—with pvalues fine-tuned via Kenward-Roger method.00 5.22814 30 20 10 5 1 0 0.  Model should be significant based on F-test: • P-value < 0.45 t-Value Limit 2. until all other effects fall below the Bonferroni and/or t-value limit.10.82734 2.79 99 AC 7.05: significant . Verify the ANOVA assumptions by looking at the residual plots.Rev 2/26/14 Factorial Analysis Guide (p1/2) 1.) a) Review the ANOVA results. b) Examine the F tests on the regression coefficients.22 2 3 4 5 6 7 8 9 10 11 12 13 14 15 21. Use half-normal probability plot to select model. Compute effects.63 Rank |Effect| 2. Look for terms that can be eliminated. • P-value > 0. • P-value > 0.

20 28.00 35.00 24.40 26.20 28.00 92.00 A: Temperature (deg C) A: Temperature (deg C) c) Cube plot (especially useful if three factors are significant): Cube Filtration Rate (gallons/hr) 72.00 24.00 A+: 35.80 24.00 2.60 32.40 30.00 D-: 15.25 C: Concentration (percent) C+: 4.20 28.50 3.25 74.60 A: Temperature (deg C) 28.00 3.00 26.00 C: Concentration (percent) A: Temperature (deg C) 2-2 2.50 70 3.00 F iltration R ate (g allons/hr) C: Concentration (percent) 90 3.00 32.00 A: Temperature (deg C) d) Contour plot and 3D surface plot: Filtration Rate (gallons/hr) 4.625 D: Stir Rate (rpm) C-: 2.375 61.40 30.125 44.Rev 2/26/14 4.00 26.80 30.40 30.00 .375 A-: 24.60 32. Explore the region of interest: a) One Factor plot (don’t use for factors involved in interactions): b) Interaction plot (with 95% Least Significant Difference (LSD) bars): Interaction One Factor Warning! Factor involved in multiple interactions.50 80 70 60 50 40 50 4.50 35.60 32.25 69.00 80 60 2. 110 100 Filtration Rate (gallons/hr) 100 Filtration Rate (gallons/hr) C: Concentration (percent) 110 90 80 70 60 90 80 C+ 70 60 50 50 40 40 C- 24.20 2.00 46.80 35.25 D+: 30.00 100.00 26.80 35.

10). Replicate runs to estimate pure error (needed to statistically assess the lack of fit). rather than just the additional terms for that order model as in the sequential table.10..can the complexity of the polynomial be reduced? Look for terms that can be eliminated. c) Check for “Adeq Precision” > 4. the design lacks: i. b) Examine the F tests on the regression coefficients . 2-3 . i.e.2 below the "Adj R-Squared". This is a signal to noise ratio given by the following formula: LMmaxeY j − mineY j OP MM VeY j PP > 4 N Q p pσ 2 1 n V Y =  V Y = n n i=1 ej ej = number of model parameters (including intercept (b0) and any block coefficients) σ2 = residual MS from ANOVA table n = number of experiments d) Check that "Pred R-Squared" falls no more than 0. d) “Model Summary Statistics”: Focus on the model with high “Adjusted R-Squared” and high “Predicted R-Squared”. Be sure to maintain hierarchy.05  p-value > 0. If so. p-value < 0.10  No lack-of-fit reported? If so. The F-test is for the complete model. these can not be selected. coefficients having p-values > 0. 2.Rev 2/26/14 Response Surface/Mixture Analysis Guide (p1/3) 1.10  c) “Lack of Fit Tests”: Want the selected model to have insignificant lack-of-fit. b) “Sequential Model Sum of Squares”: Select the highest order polynomial where the additional terms are significant and the model is not aliased. Check the selected model: a) Review the ANOVA. consider model reduction. Excess unique points beyond the number of model terms (to estimate variation about fitted surface). p-value < 0. and/or ii.05  p-value > 0. Model should be significant (p-value < 0.05) and lack-of-fit insignificant (P-value > 0. Select a model: a) “WARNING”: Note which models are aliased.

0 B C A B A C 60. 4. Refer to the Residual Analysis and Diagnostics Plots Guide.000 Deviation from Reference Point (Coded Units)  Mixture trace plot (view Piepel’s direction for broadest paths) X1 Trace (Piepel) 160 90 C 140 Viscosity (mPa-sec) 70 50 10 B 80 C 60 30 30 10 100 AB A 40 30 50 50 120 20 70 70 -0.  RSM perturbation plot Perturbation 100.800 90 90 10 X3 Deviation from Reference Blend (L_Pseudo Units) Perturbation/Trace plots are particularly useful after finding optimal points.400 0.200 0. Explore the region of interest: a) Perturbation/Trace plots to choose the factor(s)/component(s) to “slice” through the design space. In the RSM and mixture examples below.0 Conversion (%) 90.000 0. 2-4 .0 80.0 70.500 1. take slices of factor “A”.000 -0. Verify the ANOVA assumptions by looking at the residual plots.0 -1.600 0.200 0.0 50. Choose ones having small effects (flat response curve) or components having linear effects (straight). They show how sensitive the optimum is to changes in each factor or component.000 0.500 0.400 X2 -0.Rev 2/26/14 3.

00 42.80 2.00 40.80 100 C: catalyst (%) 85 90 2.000 4.000 C: Urea (%) Viscosity (mPa-sec) A (5. slicing on the factors/components identified from the perturbation/trace plots as well as any categorical factors. Perform “Numerical” optimization to identify most desirable factor (component) levels for single or multiple responses.00 70 46.00 A: time (min.00 C: catalyst (%)  Mix A: Water (%) 5.60 2.000 B: Alcohol (%) C (2.) 44. See Point Prediction for prediction interval (PI) expected for individual confirmation runs.00 46.40 Conversion (%) 6 80 75 2. 2-5 .000 160 140 120 100 80 60 40 20 B (4.000) 2 60 80 80 100 4.00 50.00 44.00 A: time (min. See Optimization Guide for details.00 90 75 2.00 50 48.000) 120 2 2 3.60 2.000) A (3.000 60 Viscosity (mPa-sec) 2 2.40 42.000 40 2.000) 5.Rev 2/26/14 b) Contour plots to explore your design space.) 2.20 2.00 40.00 65 60 2.00 48.000) B (2. 6.  RSM Conversion (%) 3.20 80 70 60 50.000) C (4.00 3. View the feasible window (‘sweet spot’) via “Graphical” optimization (‘overlay’ plot).00 2.

0001) ♦ L process terms are a significant addition ([8] p<0.54010) ♦ Add L process to significantly improve ([14] p<0. the QxL combined mixture-by-process model is suggested. it’s the one with the highest adjusted and predicted R-squared (row [14] 0. For example.9883) ♦ The next two models (rows [4] and [5]). to eliminate insignificant terms. ♦ The next two models ([10] and [11]).9601 and 0.0001) Thus.0001 in row [2]) ♦ Adding 2FI process terms provide no benefit ([3] p=0. in this case: quadratic mixture by linear process (QxL). try model reduction.0001) ♦ 2FI process do not add significantly ([9] p=0. 2-6 . [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] This sequential table shows the significance of terms added layer-bylayer to the model above. LxQ and LxC are aliased ♦ From LxM [12] add Q Mix: No help here ([13] p=0.6237). ♦ Linear (L) mixture terms are significant ([7] p<0.Rev 2/26/14 Combined Mixture/Process Analysis Guide 1. as in this case. in this case starting with the mean by mean (MxM) model (row [1]): ♦ Linear (L) process terms provide significant information beyond the mean (M) model (p<0. MxQ and MxC are aliased – do not pick them! Start again from MxM (row [6]). Often. preferably backward. Due to the complexity of combined models.9240). Select a model: Look for what’s suggested in the Fit Summary table. 2.

00 -4.00 -2. Residuals vs.20 14. Other diagnostic plots may provide interesting information in some situations. 99 99 95 95 90 90 80 80 70 70 50 50 30 30 20 20 10 10 5 5 1 1 -2.00 0.00 Predicted 0.60 0.00 0.73 0.00 -2.00 10.00 0.00 2.00 4.58 2. Diagnostic plots 1.00 Predicted BAD: Megaphone shape GOOD: Random scatter 2-7 1.14 -0. ALWAYS review these plots! A.should be random scatter.40 0.00 8.20 .60 -0.80 1.34 1.14 1.00 0.88 GOOD: Linear or Normal b) Residuals (ei) vs predicted .90 0.00 -4.81 BAD: S shape -1. Plot the (externally) studentized residuals: a) Normal plot . Predicted 4.Rev 2/26/14 Residual Analysis and Diagnostic Plots Guide (p1/5) Residual analysis is necessary to confirm that the assumptions for the ANOVA are met.00 6. Predicted Residuals vs.01 1.00 2.should be straight line.00 6.00 4.00 12.00 -6.00 2.

00 -4. Run 6.56 36.00 1 4 7 10 13 1 16 4 7 10 13 16 Run Number Run Number BAD: Trend GOOD: No pattern ♦ a problem with the model. Run Residuals vs.81 49. Residuals vs.00 1 4 7 10 13 16 Run Number 90.00 -2.80 89.00 0.00 Externally Studentized Residuals Also. Groups of points above or below the line indicate areas of over or under prediction.Rev 2/26/14 c) Residuals (ei) vs run .00 87. Run 4.00 36.should be random scatter.50 . ♦ a special cause that merits ignoring the result or run. look for externally studentized residuals outside limits.00 0.80 50.00 2.31 76.22 61.00 2.00 61.50 74.00 -4. Residuals vs.50 48.25 Predicted Predicted 2.00 2.65 74.00 0.00 -2.00 -4. no trends.08 Actual Actual Poor Prediction Better Prediction 2-8 86.00 -2.50 49.00 76. View the predicted vs actual plot whose points should be randomly scattered along the 45 degree line. These runs are statistical outliers that may indicate: 4. ♦ a transformation.00 Externally Studentized Residuals Externally Studentized Residuals 4.00 36.75 36. 63.05 63.00 -6.

60 1.00 Block GOOD: Random scatter both ends 2-9 .00 2.40 1. Should be split by the zero-line at either end of the range – no obvious main effect (up or down). Watch ONLY for very large differences.00 -2. Block 4.00 1.80 2.20 1.85 7.00 0. in which case the recommendation will be “None. If you see an effect.94 1. Residuals vs.20 1. Red lines indicate a 95% confidence interval associated with the best lambda value. Stat-Ease software recommends the standard transformation.40 1.41 5.43 2. go back. such as log.00 Externally Studentized Residuals Externally Studentized Residuals 4.” 3.00 -4.00 -4.60 1.Rev 2/26/14 6. Use the Box-Cox plot (not available for split-plot designs) to determine if a power law transformation might be appropriate for your data.00 1.00 Block BAD: More variation at one end 1. The blue line indicates the current transformation (at Lambda =1 for none) and the green line indicates the best lambda value.00 1.50 2.80 2. Block Residuals vs. closest to the best lambda value unless the confidence interval includes 1.40 5.05 1.00 -2. add it to the predictive model and assess its statistical significance. Relatively similar variation between levels.92 Ln(ResidualSS) Ln(ResidualSS) 3.95 4.00 2. Residuals (ei) vs factor – especially useful with blocks.00 0.44 -3 -2 -1 0 1 2 -3 3 -2 -1 0 1 2 3 Lambda Lambda Before Transformation After Transformation 4.

5000 0. These runs will unduly influence at least one model parameter.75 0.2500 0. Investigate the run with the largest Cook’s Distance first. These runs will be fitted exactly with no residual! 1. Often. Caution: For small designs this rule-of-thumb may be overly sensitive. If identified prior to running the experiment.0000 1 6 11 16 21 26 1 Run Num ber 3 5 7 9 11 13 15 17 Run Number BAD: Some at twice the average GOOD: All the same 3.7500 Leverage Leverage g 1. 2-10 .00 0. Be especially careful of any leverages at one (1.75 0. Otherwise all you can do is check the actual responses to be sure they are as expected for the factor settings. Watch for values outside of ±2. it can be replicated to reduce leverage. Cook’s Distance (not available for split-plot designs) helps if you see more than one outlier in other diagnostic plots. and seeing how this affects the model fit. so do not become alarmed at seeing a number of points out of limits. a) DFFITS (difference in fits) is another statistic helpful for detecting influential runs.Rev 2/26/14 Influence plots 1.0000 0.0). other apparent outliers can be explained by the model.25 0. Watch for leverage vs run values at or beyond twice the average leverage.00 1 4 7 10 13 16 Run Num ber 2.50 0.00 0. 1.50 0. Deletion diagnostics – statistics (not available for split-plot designs) calculated by taking each run out. one after the other.00 Cook's Distance 0. if this run is ignored due to a special cause.25 0.

4.Rev 2/26/14 b) DFBETAS (difference in beta coefficients) breaks down the impact of any given run on a particular model term. consider whether it falls beyond a reasonable range (for example.00 1 6 11 16 21 26 31 Run Num ber Statistical Details on Diagnostic Measures Residual ( ei=yi -y ˆi ): Difference between the actual individual value ( yi ) and the value predicted from the model ( ˆ yi ). A leverage of 1 means the predicted value at that particular case will exactly equal the observed value of the experiment (residual=0. ( ) −1 Leverage ( hii = x iT X T X x i where x is factor level and X is design matrix): Numerical value between 0 and 1 that indicates the potential for a case to influence the model fit. Runs with large t values (rule-of-thumb: |t| > 3. The maximum leverage an experiment can have is 1/k. Internally Studentized Residual ( ri = ei s (1 − hii ) ): The residual divided by the estimated standard deviation of that residual (dependent on leverage).) The sum of leverage values across all cases (design points) equals the number of coefficients (including the constant) fit by the model. if so.20 1. it may be that an axial (star) point in a CCD projects outside of the feasible operating region) and. It represents the number of standard deviations between this predicted value and the actual response.93 DFBETAS for C 3. If you see an DFBETAS for C vs. Externally Studentized Residual ( ti = ei s−1 1 − hii ): This “outlier t” value is calculated by leaving the run in question out of the analysis and estimating the response from the remaining runs.46 -0. try ignoring this particular run.5) are flagged and should be investigated. where k is the number of times the experiment is replicated. Run excessive value for a particular factor. Values larger than 2 times the average leverage are flagged.27 -2. 2-11 . which measures the number of standard deviations separating the actual from predicted values.

DFBETAS ( DFBETASj. It represents the number of standard errors that the jth beta-coefficient changes if the ith observation is removed. this statistic becomes undefined for leverages of one (h=1). Myers. 3rd edition. or a design point far from the others. Inc. DFFITS represents the externally studentized residual (ti) magnified by high leverage points and shrunk by low leverage points.−i s−1 c jj . an incorrect model. so make sure to use the pull-down menu and click through the terms (the down arrow is a good shortcut key – also. Mathematically it is the studentized difference between the predicted value with and without observation “i”. 2nd edition. Cases with large Di values relative to the other cases should be investigated. References: 1. Weisberg. alternatively DFFITSi = ti  ii  hii 1 − hii  s−1 ): DFFFITS measures the influence each individual case (i) has on the predicted value (see Myers2 page 284. As shown by the alternative formula. Stanford: Applied Linear Regression.Rev 2/26/14 Note: For verification runs.−i  h  . 2005. 2. John Wiley & Sons. Note that DFFITs becomes undefined for leverages of one (h=1). try the wheel if you have one on your mouse). Like DFFITS. Look for mistakes in recording. Cook's Distance ( Di = 1 2  hii  ri   ): p  1 − hii  A measure of how much the regression would change if the case is omitted from the analysis (see Weisberg1 page 118). Duxbury Press.i = βj − ˆ y j.) It is a function of leverage (h). cjj is the jth diagonal element of (X’X)-1): DFBETAS measures the influence each individual case (i) has on each model coefficient (βj). Relatively large values are associated with cases with high leverage and large studentized residuals. 2000. 2-12 . DFFITS ( DFFITSi = 1/2 ˆ yi − ˆ yi. DFBETAS are calculated for each beta-coefficient. Raymond: Classical and Modern Regression with Applications. this is calculated using the predicted error (an expansion of the hat matrix).

“target”. Factors-only (default “in range”): “equal-to”. ♦ Bar Graph displays how well each variable satisfied their criterion..e. Cycle through rank of solution from top to bottom. Run the optimization (press Solutions).25 0. Responses-only: “none” (default). greater than 4. response values. Set the following criteria for the desirability optimization: a) Goal: “maximize”. c) Weight (optional): Enter 0. ♦ Report shows settings of the factors. 0. ♦ Ramps show settings for all factors and the resulting predicted values for responses and where these fall on their desirability ramps.e.75 4. d) Importance (optional): Changes goal’s importance less (+) to more (+++++) relative to the others (default +++). b) Insignificant lack-of-fit. The default of 1 keeps it linear. i.00 90 B: temperature85 80 40 45 A: time 2-13 50 . d) Well-behaved residuals.1 to 10 or drag the desirability ramp up (lighter) or down (heavier).50 0. Check for: a) A significant model. Be sure the fitted surface adequately represents your process. Analyze each response separately and establish an appropriate transformation and model for each. 3. and desirability for each solution from top to bottom. a large F-value with p<0.e. “minimize”. i. 2. Graph the desirability. b) Limits lower and upper: Both ends required to establish the desirability from 0 or 1.10. c) Adequate precision. Weights >1 give more emphasis to the goal and vice-versa.. “in-range” and “Cpk”.. an F-value near one with p>0.Rev 2/26/14 Optimization Guide (p1/2) Numerical Optimization: 1.05. i. Desirability 0.

multiple passes are needed to find the “best” factor levels to simultaneously satisfy all operational constraints.65 2. 2.0 2.0 2.00 A: time Suggestions for achieving desirable outcome: Numerical optimization provides powerful insights when combined with graphical analysis.7398 63.Rev 2/26/14 Graphical Optimization: 1.75 C: catalyst ♦ Upper only if minimized Overlay Plot 3.50 2. It overlays all responses – shaded areas do not meet the specified criteria. Conv ersion: Conversion: Activity: X1 X2 80.00 47.00 Activ ity : 60. 2-14 . it cannot substitute for subject matter knowledge.0 89. However.0 2.25 C onv ersion: 80. set broad optimization criteria and then narrow down as you gain knowledge. Most often.50 45. Graph the optimal point identified in the numerical optimization by clicking the #1 solution.50 50. For example. only to find zero desirability everywhere! To avoid finding no optimums.0005 46. The flagged window shows the “sweet spot”.76 Activ ity : 66. you may define what you consider to be optimum.00 42. Criteria require at least one limit for at least one response: ♦ Lower only if maximized (unlike numerical optimization where you must enter both lower and upper limits!) ♦ Lower and upper (specification range) if goal is target.00 40.

1st & 2nd Derivatives of Transformations Transform Square root counts Loge variation Log10 variation Inverse square root Power (lambda) 0.Rev 2/26/14 Inverses.5 Formula y′= y+k y′ = ln ( y + k ) y′ = log ( y + k ) 2 y′ y′ 1 y′= y+k y = y′ −2 Inverse y = y′ − k y = e −k y = 10 − k 1st Derivative ∂y = 2y′ ∂y′ ∂y = ln ( e ) ey′ = ey′ ′ ∂y ∂y = ln (10 ) 10y′ ′ ∂y ∂y = −2y′−3 ′ ∂y 2nd Derivative ∂2y =2 ∂y '2 ∂2y = ey ' 2 ∂y ' 2 ∂2y = ( ln (10 ) ) 10 y ' 2 ∂y ' ∂2 y = 6y '−4 2 ∂y ' ArcSin Square Root Binomial data y is a fraction (0-1) y’ in radians Logit Asymptotically bounded data LL=lower limit UL=upper limit NA NA y′=sin-1 y  y − LL  y′=ln    UL − y  Transform Inverse Rates Power when all else fails Power (lambda) -1 λ Formula y′= 1 y+k y′= ( y+k ) λ 1 y=( sin ( y′ ))2 Inverse y = y′−1 − k y = ( y′ ) λ + k 1st Derivative ∂y = − y ′ −2 ∂y′ 1  ∂y 1 −1 = ( y′ ) λ  ∂y′ λ 2nd Derivative ∂2 y = 2y '−3 ∂y '2 1  ∂2y 1  1  −2 = − 1 y ' ( ) λ    2 ∂y' λλ  2-15 ∂y = 2 sin ( y′ ) cos ( y′ ) ∂y′ ∂2y = 2cos ( 2y ' ) ∂y'2 y= −k ( ) UL ey′ + LL 1 + ey ′ ey′ (UL − LL ) ∂y = 2 ∂y′ 1 + ey ′ ( ( ) ) y' y' ∂ 2 y e 1 − e (UL − LL ) = 3 ∂y'2 1 + ey ' ( ) .5 0 0 -0.

Rev 2/26/14

This page left blank intentionally as a spacer.

2-16

Rev 2/26/14

Section 3:
Appendix

Rev 2/26/14

This page left blank intentionally as a spacer.

2709 0.0392 0.3409 0.0087 0.8 2.0281 0.1075 0.0174 0.0015 0.0001 0.0336 0.0004 0.1314 0.0143 0.0010 0.0001 0.0014 3.4721 0.0009 0.0000 0.2119 0.4840 0.0885 0.4129 0.0005 0.3085 0.2327 0.1190 0.0030 0.1292 0.0020 0.5 0.0004 0.0003 0.0005 0.5000 0.0516 0.0000 0.0084 0.2743 0.0170 0.4801 0.0066 0.0764 0.0001 0.2005 0.0001 0.0329 0.3228 0.06 0.3520 0.0001 0.6 1.0 0.2912 0.5 1.0035 0.0838 0.1685 0.0808 0.2810 0.2643 0.0000 0.0132 0.0571 0.8 3.0359 0.0119 0.0002 0.0384 0.02 0.0 0.0001 0.0869 0.0006 0.0233 2.1949 0.0274 0.3669 0.0002 0.0001 0.0008 0.0001 0.0643 0.1003 0.7 3.5 3.0116 0.1611 1.1 0.8 0.0094 0.0008 0.0003 0.1401 0.4207 0.0000 3-1 .4168 0.2 1.0045 0.0002 0.4052 0.0294 0.0188 0.0064 0.0110 0.0026 0.0057 0.3821 0.0008 0.0268 0.0287 0.0465 0.0002 0.0418 0.0034 0.0039 0.0000 0.2061 0.0003 0.4761 0.1423 0.08 0.0314 0.2 3.6 0.0192 0.0032 0.3783 0.0505 0.0582 0.0002 0.6 3.9 0.0003 0.0049 0.7 2.0003 0.2946 0.1762 0.0003 0.1814 0.3557 0.0002 0.4681 0.0001 0.0217 0.0749 0.0013 0.0010 0.0038 0.0526 0.0091 0.3300 0.0002 0.0001 0.0344 0.0000 0.4 1.0212 0.0594 0.0001 0.0048 0.2546 0.4562 0.1841 0.0197 0.0475 0.0001 0.0005 0.0069 0.1170 0.0025 0.1977 0.0036 0.1093 0.3 3.0055 0.1894 0.0019 0.0001 0.4602 0.0060 0.0001 0.7 0.0351 0.0708 0.1922 0.0004 0.0207 0.0001 0.3707 0.0250 0.4 0.0001 0.0537 0.0655 0.05 0.0011 0.4 3.0096 0.7 1.2843 0.0002 0.1151 0.0023 0.0007 0.0102 0.0668 0.3192 0.0016 0.0007 0.1492 0.0052 0.2420 0.Rev 2/26/14 Z-Table: Tail area of unit normal distribution z 0.0024 0.0009 0.0006 0.0054 0.0001 0.3050 0.0000 0.0021 0.0023 0.2033 0.2389 0.0040 0.4880 0.3974 0.1711 0.0901 0.4443 0.3372 0.6 2.0006 0.4483 0.0002 0.0409 0.0000 0.3859 0.0017 0.2 0.0113 0.0009 0.0005 0.2578 0.0367 0.9 0.0003 0.0630 0.0918 0.1335 0.0027 0.00 0.0075 0.1 3.0047 0.4 2.3483 0.01 0.0001 0.0068 0.3336 0.0154 0.2266 0.2090 0.0062 0.4364 0.1 1.0089 0.1357 0.2514 0.3936 0.0721 0.0007 0.1515 0.3 1.0002 0.0162 0.1587 0.0012 0.0001 0.4286 0.0043 0.0455 0.0037 0.4325 0.0021 0.0166 0.9 0.0606 0.0099 0.1112 0.0146 0.3264 0.0015 0.9 0.4013 0.3 2.1867 0.2177 0.0150 0.2483 0.3632 0.1210 0.4960 0.0004 0.0202 0.0 0.4920 0.2148 0.1038 0.3446 0.0044 0.0136 0.0016 0.1635 0.0013 0.0139 0.1230 0.0256 0.0436 0.0853 0.1271 0.0618 0.0001 0.04 0.1020 0.2451 0.0041 0.0322 0.3121 0.0 0.1539 0.0778 0.3 0.07 0.3015 0.2 2.0001 0.0001 0.3897 0.1379 1.0985 0.09 0.03 0.8 1.2877 0.1469 0.3745 0.0222 0.0029 0.0559 0.2236 0.0951 0.0239 0.1131 0.1562 0.0003 0.0375 0.4247 0.0104 0.0446 0.5 2.2776 0.0011 0.0129 0.4641 0.0071 0.0005 0.0001 0.0001 0.0301 0.2676 0.0934 0.0968 0.0004 0.0307 0.0078 0.0495 0.0005 0.0018 0.0007 0.2358 0.0028 0.0022 0.0014 0.4404 0.0004 0.4090 0.0026 0.0011 0.0694 0.0823 0.0012 0.0548 0.0003 0.0059 0.0107 0.0013 0.0122 0.0244 0.0262 0.0073 0.0735 0.0427 0.0000 0.1660 0.0001 0.0485 0.0158 0.0401 0.2611 0.0179 0.2206 0.0681 0.0000 0.0125 0.0003 0.3594 0.0002 0.1788 0.0001 0.0002 0.0018 0.0082 0.0006 0.1056 0.1 2.0793 0.0001 0.0019 0.2981 0.0080 0.0010 3.0031 0.0008 0.4522 0.2296 0.0051 0.0006 0.0228 0.3156 0.1736 0.0033 0.1446 0.1251 0.0002 0.0183 2.0001 0.

337 1.508 2.771 2.708 1.145 2.787 2.313 1.144 4.091 3.052 2.326 63.685 0.684 0.925 5.685 0.353 2.314 2.143 2.296 1.333 1.271 0.267 0.435 3.898 2.711 1.255 0.101 2.153 3.078 1.476 1.581 3.256 0.120 2.029 3.734 1.807 318.257 0.160 3.733 3.174 3.067 3.104 3.257 0.25 0.690 0.528 2.258 0.169 3.80 0.041 4.62 31.860 1.160 2.745 3.40 0.356 1.924 8.721 1.318 4.256 0.182 2.706 0.291 df=1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 60 120 ∞ 3-2 .32 14.450 3.947 2.358 2.415 1.697 1.437 4.598 12.259 0.257 0.326 10.055 3.692 0.683 0.421 3.025 0.718 2.258 0.310 1.657 9.080 2.812 1.691 0.965 3.674 3.316 1.326 3.977 2.173 5.197 3.833 1.960 31.254 0.701 1.021 2.259 0.253 1.505 3.10 0.610 3.646 3.132 2.060 2.277 0.604 4.303 3.307 3.345 1.467 3.497 3.250 3.222 3.943 1.869 5.527 3.056 2.624 2.057 3.746 1.363 1.533 1.703 0.20 0.303 1.971 2.763 2.819 2.699 1.492 2.0005 2-tail 0.050 0.341 1.684 0.256 0.552 2.729 1.069 2.01 0.355 3.457 2.045 2.796 1.140 4.321 1.965 4.816 0.518 2.262 2.256 0.761 1.325 0.896 2.390 2.073 4.689 0.681 0.263 0.741 0.646 3.208 4.686 0.048 2.314 1.617 2.135 3.684 0.001 0.408 3.659 3.256 0.257 0.686 0.365 2.Rev 2/26/14 One-tailed / Two-tailed t-Table Probability points of the t-distribution with df degrees of freedom t tail area probability 1-tail 0.032 3.0025 0.576 127.767 3.350 1.047 3.428 3.282 6.552 3.0010 0.658 1.893 5.106 3.317 4.397 1.473 2.610 6.015 3.779 2.201 2.254 0.232 3.252 3.010 0.704 2.765 0.688 0.660 2.31 22.687 0.262 0.423 2.467 2.213 7.690 3.920 2.372 1.707 3.440 1.551 3.315 1.000 0.000 1.567 2.807 2.583 2.460 3.330 1.684 1.776 2.286 3.256 0.015 1.707 3.325 1.074 2.703 1.396 3.289 0.539 2.385 3.256 0.328 1.695 0.260 0.645 12.980 1.002 0.462 2.258 0.677 0.679 0.265 0.541 3.819 3.500 2.771 1.718 0.821 2.850 3.485 3.860 2.479 2.930 3.289 1.922 3.05 0.878 2.861 2.711 0.781 4.650 2.02 0.727 0.042 2.638 1.093 2.886 1.005 0.571 2.228 2.10 0.408 5.787 3.089 7.683 0.671 1.747 3.845 2.821 6.852 3.038 3.773 4.841 4.261 0.257 0.110 2.598 4.579 3.064 2.260 0.717 1.078 3.50 0.383 1.750 2.674 3.706 4.740 1.782 1.012 2.764 2.694 0.683 0.959 5.681 2.179 2.725 1.221 4.297 4.700 0.453 5.833 3.030 2.365 3.131 2.714 1.119 3.318 1.686 3.311 1.323 1.690 3.090 636.921 2.883 3.499 3.256 0.086 2.792 3.797 2.915 2.688 0.756 2.998 2.319 1.256 0.306 2.785 4.831 2.447 2.372 3.485 2.587 4.725 3.373 3.697 0.602 2.753 1.895 1.0050 0.025 3.501 4.706 1.

71 34.65 12.09 16.554 0.297 0.14 31.26 9.33 99.57 38.75 22.26 6.34 42.53 32.68 11.45 16.81 6.64 42.26 7.92 43.73 3.05 3.54 10.85 11.831 1.46 45.39 0.60 3.34 39.44 9.455 1.92 0.41 32.36 14.34 19.14 22.74 60.32 128.67 9.69 29.01 33.48 33.63 3.84 9.25 19.17 74.56 30.83 3.99 17.711 0.50 25.66 16.14 129.34 15.96 23.77 55.575 1.75 18.13 98.27 35.34 29.13 77.59 10 11 12 13 14 2.52 11.76 43.010 0.52 12.66 23.072 0.30 27.80 45.63 118.98 11.33 89.26 5.49 21.14 19.70 6.36 11.56 15.20 12.51 79.86 11.93 17.28 51.81 21.65 74.91 9.48 20.96 8.34 12.90 9.56 46.02 15.76 57.75 0.60 12.29 18.34 23.975 0.05 16.71 27.94 19.872 1.22 7.07 15.59 28.42 83.58 8.36 40.72 46.61 6.55 20.84 21.56 8.34 53.51 16.26 51.33 5.31 19.03 8.020 0.16 11.62 56.051 0.14 25.39 2.05 37.11 18.22 116.34 30.20 10.21 1.81 12.14 0.39 69.22 27.51 34.989 1.17 59.412 0.24 1.40 42.58 40.67 16.584 1.07 5.34 26.95 22.35 5.29 49.49 26.34 59.34 8.79 6.28 10.15 65.01 6.61 2.115 0.78 38.34 25.46 55.94 26.14 5.54 24.68 14.85 34.01 5.01 17.20 44.39 9.27 6.49 0.17 9.65 50.74 0.56 3.47 17.30 140.98 44.102 0.34 20.88 13.12 15.49 4.70 71.09 10.484 0.66 42.87 30.20 34.19 26.89 63.36 0.23 5.44 13.64 46.24 10.31 23.06 0.10 2.65 41.20 2.17 2.28 49.29 82.29 61.34 27.31 45.95 104.98 77.24 1.15 1.39 10.43 32.40 85.70 14.65 12.59 10.16 2.995 0.34 10.82 9.35 13.24 29.77 4.58 32.20 67.82 4.48 48.57 15.89 10.63 8.016 0.005 7.37 20.33 18.09 21.34 12.35 7.34 21.34 11.02 106.53 96.02 7.16 29.92 39.55 19.78 0.34 71.54 19.99 27.34 17.61 15.81 32.60 5.11 5.94 20.17 36.64 2.09 0.85 15.17 35.12 13.30 29.24 18.38 9.15 88.99 7.90 4.34 17.49 91.30 7.43 112.33 64.81 32.91 7.40 13.84 7.Rev 2/26/14 Cumulative Distribution of Chi-Square Area = α Probability of a Greater value df 0.89 6.20 28.68 21.66 3.21 24.92 12.67 66.32 2.96 48.65 2.59 14.83 14.53 101.89 40.73 26.68 3.21 11.34 24.57 4.88 113.88 10.15 16.62 33.63 90.99 35.48 36.34 18.49 28.76 67.57 118.94 52.44 53.40 5.12 23.103 0.70 1.71 37.09 13.12 135.34 13.55 9.81 36.19 37.98 59.87 11.84 14.58 6.352 0.207 5 6 7 8 9 2 χ 2 χα 0.69 12.05 3.58 5.11 4.34 69.41 34.004 0.04 10.32 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 50 60 70 80 90 100 4.82 31.79 13.42 37.57 4.69 2.35 6.23 5.41 29.59 50.28 73.93 8.72 37.02 7.07 12.34 28.79 20.67 33.16 38.25 3.90 0.33 1.44 31.43 8.87 5.08 39.676 0.38 100.31 16.22 11.30 95.86 0.71 18.58 88.37 3.99 52.35 11.98 17.36 40.63 7.84 14.77 79.03 22.02 13.15 80.54 61.93 40.34 6.30 10.18 2.45 16.85 15.36 23.92 23.77 25.28 21.64 9.04 27.79 24.74 7.56 36.14 28.81 18.11 41.65 38.23 8.19 51.17 2.09 40.13 14.38 16.34 1.49 45.94 4.60 29.81 63.17 3-3 .80 44.50 0.75 70.48 21.33 79.95 0.80 34.31 10.33 66.33 124.001 0.41 7.025 5.61 22.04 19.91 12.00 26.58 107.04 14.59 12.25 7.76 28.08 90.34 22.34 13.71 4.46 13.06 18.84 5.62 30.07 6.07 3.29 41.55 13.24 14.57 14.74 37.53 32.85 14.69 20.12 13.08 2.56 43.04 7.15 124.28 0.34 27.00 41.25 1.53 43.57 24.57 4.12 10.99 1 2 3 4 0.77 20.83 24.34 24.92 35.69 46.26 14.65 109.34 16.04 11.28 18.18 45.26 8.38 35.85 30.64 9.72 23.19 31.216 0.46 4.34 49.00 33.74 26.69 76.93 48.211 0.

684 1.041 1.035 1.158 2.075 2.811 1.487 1.685 29 2.122 2.949 2.214 2.759 23 2.887 2.327 2.010 3.317 2.718 26 2.038 2.307 2.953 1.451 2.105 2.863 49.996 1.441 3 5.136 2.416 2.005 1.703 2.706 27 2.937 2.730 25 2.293 9.511 2.780 3.595 8 3.088 2.055 2.142 2.273 2.740 1.543 120 2.707 1.325 4.606 2.047 1.482 100K 2.948 1.107 4.819 1.073 2.967 1.706 2.785 2.909 2.026 2.244 2.230 5.495 2.827 1.814 20 2.130 2.632 1.920 3.005 1.875 1.857 1.874 1.927 2.984 1.461 2.500 53.014 2.624 2.368 3.234 2.995 1.380 2.940 1.051 4.299 2.956 1.858 60.982 1.381 9.873 1.244 2.219 2.057 1.806 2.551 2.924 2.000 9.128 2.283 2.195 2.695 2.245 2.347 2.731 1.866 1.979 3.943 1.575 2.303 2.717 1.109 2.748 2.652 1.836 7 3.827 1.207 2.860 2.339 3.091 1.845 1.074 2.660 2.722 1.149 2.220 61.775 1.138 2.102 2.933 1.393 2.977 1.593 55.053 2.469 2.007 14 3.391 5.233 2.437 2.794 21 2.178 2.855 1.668 2.283 2.167 2.207 6 3.924 2.829 1.662 1.010 1.941 1.283 2.331 2.796 1.243 2.752 2.645 2.668 2.519 2.972 1.538 5.040 1.055 3.503 2.421 K (Multiply this value by 1000) 3-4 .064 1.927 1.807 2.184 4 4.165 2.894 2.847 1.983 1.248 2.367 9.365 2.405 3.462 2.257 3.920 1.188 2.999 1.162 9.238 3.906 1.919 1.008 1.561 2.884 1.416 2.Rev 2/26/14 F-Table for 10% Percentage points of the F-distribution: upper 10% points df num 1 2 3 4 5 6 10% F 7 8 9 10 15 20 df den 1 39.980 1.196 2.896 1.827 2.128 2.060 2.082 2.632 2.339 2.906 59.744 24 2.079 2.900 1.871 2.028 2.309 5.326 9.360 3.130 1.726 2.340 2.844 5 4.480 2.017 1.226 2.605 2.836 1.522 2.249 2.536 2.395 2.971 1.157 2.693 2.740 2 8.440 2.243 9.392 9.760 1.738 1.783 1.193 2.545 1.892 1.890 1.440 2.670 1.142 2.289 3.924 16 3.589 2.624 2.453 3.774 1.115 2.918 2.048 2.201 11 3.298 10 3.599 1.912 1.722 1.086 2.833 57.904 1.990 2.901 2.549 2.342 2.865 1.936 3.561 2.361 2.154 2.200 5.092 2.877 1.158 2.526 9.749 1.439 59.520 3.776 22 2.164 2.763 2.728 2.463 3.333 2.204 58.285 2.589 2.935 1.113 2.023 1.425 9 3.218 2.291 2.606 2.225 2.174 2.909 1.323 2.007 2.824 1.458 3.347 2.123 12 3.695 28 2.208 2.425 9.001 1.983 2.490 2.266 2.891 17 3.776 3.489 2.946 1.351 2.119 2.819 1.177 2.060 13 3.619 3.965 1.763 1.522 2.394 2.184 2.434 2.837 19 2.191 4.061 2.240 5.049 1.895 1.538 2.058 2.883 2.726 2.195 2.024 1.397 2.771 1.285 5.881 2.316 3.505 2.103 2.791 2.152 2.307 2.176 2.102 2.389 2.276 2.308 2.059 1.962 15 3.073 2.605 60 2.937 1.955 3.181 3.997 1.252 5.845 1.870 3.676 30 2.611 2.988 1.091 2.177 2.849 1.286 2.919 1.929 1.961 2.961 2.464 2.545 4.377 2.274 2.462 5.414 2.862 18 3.767 1.240 58.297 3.793 1.992 1.952 1.528 2.835 2.343 5.589 3.975 2.961 1.347 2.927 1.560 2.266 5.195 61.603 1.958 2.349 9.014 1.060 3.667 40 2.887 1.028 1.945 1.095 2.108 3.084 1.884 1.813 2.725 2.304 2.865 1.538 2.937 2.006 2.

027 1.449 2.459 2.572 2.581 3.099 4.832 2.54 241.848 2.671 2.163 6.354 2.839 60 4.707 2.160 2.95 248.591 2.467 3.508 2.910 1.072 2.840 2.256 3.103 3.337 2.177 2.587 3.571 K (Multiply this value by 1000) 3-5 .866 3.242 3.446 3 10.378 2.99 236.372 2.071 23 4.646 12 4.196 3.180 2.533 2.348 2.711 2.072 1.701 2.048 24 4.544 2.207 4.464 2.661 2.587 2.097 2.450 2.239 3.297 2.250 2.087 2.605 2.525 2.328 16 4.996 2.646 2.803 5 6.854 2.591 6.447 2.901 2.932 40 4.183 3.948 2.637 3.747 3.924 1.714 2.330 19.165 2.353 19.728 2.403 3.459 4.548 2.922 2.066 3.112 2.290 2.474 2.786 5.447 2.743 2.217 3.88 240.340 2.600 3.16 233.844 3.397 2.106 2.010 1.831 1.190 2.147 4.375 2.396 19.592 3.996 2.818 4.359 2.817 2.296 19.265 2.965 4.852 2.920 3.511 3.094 6.633 3.764 2.429 19.089 2.277 9.120 3.388 15 4.599 2.545 2.726 3.739 3.938 3.357 3.617 2.321 2.096 22 4.013 8.366 2.328 2.534 4.301 3.388 6.928 2.735 4.660 4 7.368 2.71 224.269 2.690 2.236 2.409 5.001 3.388 2.965 2.776 2.709 6.958 2.234 2.326 3.117 9.254 2.538 2.842 2.896 2.960 2.108 2.628 2.608 5.591 4.621 2.640 2.028 2.741 2.513 19.197 2.015 1.320 2.318 4.887 8.482 3.099 2.291 2.Rev 2/26/14 F-Table for 5% Percentage points of the F-distribution: upper 5% points df num 1 2 3 4 5 6 5% F 7 8 9 10 15 20 df den 1 161.445 2.278 2.659 100K 3.703 8.381 3.442 2.972 3.719 2.708 3.947 2.232 2.606 2.796 2.737 4.351 3.522 3.056 2.211 2.225 3.041 5.849 2.588 2.230 3.699 2.77 238.534 2.680 2.355 2.950 4.374 3.276 17 4.577 2.179 3.978 2.420 2.758 2.056 1.544 13 4.941 8.422 3.445 8 5.411 3.549 2.885 3.772 4.214 2.385 19.602 2.810 2.325 3.223 2.490 3.191 19 4.58 230.944 6.236 2.388 3.845 8.414 3.403 2.179 3.836 1.500 3.049 2.699 2.759 2.619 4.249 2.385 2.573 2.558 2.666 1.025 2.175 2.117 4.347 4.041 1.982 3.514 2.369 2.334 2.164 19.336 2.913 2.740 2.287 3.438 3.143 4.171 3.279 3.687 3.344 3.176 2.060 3.137 3.494 3.300 2.934 2.072 2.488 2.284 4.347 3.959 1.308 2.247 19.874 7 5.01 2 18.987 5.787 3.040 1.974 28 4.012 2.027 25 4.990 27 4.210 3.50 215.661 2.552 9.839 2.641 2.634 3.167 2.371 19.494 2.256 6.677 3.072 3.266 2.192 5.786 8.007 26 4.321 2.305 2.443 3.845 2.423 2.150 9 5.151 2.352 2.945 30 4.007 2.124 21 4.490 2.456 2.432 2.346 2.000 19.260 3.682 3.603 2.767 2.293 3.939 1.451 3.050 4.773 2.863 3.510 2.387 4.667 3.006 2.544 2.999 5.077 1.220 2.155 20 4.876 4.750 1.895 2.915 2.459 14 4.204 3.812 8.098 2.936 10 4.748 120 3.858 5.218 3.085 3.373 2.959 29 4.975 2.806 3.342 2.714 2.421 2.412 2.016 1.657 2.555 3.543 3.774 11 4.790 2.558 6 5.128 2.478 3.135 3.477 2.203 2.423 2.685 2.494 2.964 5.614 2.259 3.991 2.880 1.866 2.88 245.150 2.405 2.282 2.128 9.45 199.204 2.757 4.493 3.275 2.528 2.230 18 4.316 2.020 2.838 3.009 2.993 1.393 2.796 2.095 3.127 2.255 2.753 2.124 2.463 2.

619 5.023 2.269 4.325 2.123 3.837 2.844 6.874 2.511 2.289 3.652 4.146 5.709 df den 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 60 120 ∞ 3-6 .288 8.055 3.923 2.276 2.434 7.051 3.663 3.213 2.763 2.700 2.616 2.607 3.10 38.155 3.717 5.667 2.761 4.913 2.929 2.844 2.862 2.956 2.779 2.448 17.759 3.353 3.051 3.11 948.042 5.420 4.468 4.664 3.059 5.610 5.573 2.825 1.651 2.721 3.221 3.256 5.779 3.484 4.786 2.502 3.428 5.996 3.508 4.408 3.824 2.299 2.600 4.85 937.681 2.677 2.859 3.944 1.044 3.729 2.703 2.881 3.412 2.848 2.344 2.813 8.619 4.431 39.534 2.588 5.005 2.624 2.640 2.979 7.466 2.5% Percentage points of the F-distribution: upper 2.355 39.437 2.461 4.630 4.529 2.827 5.633 5.000 39.439 15.568 4.866 2.687 4.357 4.333 3.589 3.559 2.349 4.250 3.073 7.523 5.509 2.126 3.744 2.560 4.452 2.611 2.425 2.072 3.948 2.763 2.183 3.334 2.165 39.715 5.383 4.461 4.398 39.604 3.978 5.286 3.890 5.729 3.804 3.419 3.728 3.649 8.192 8.626 3.008 2.969 2.226 3.219 3.674 2.894 2.647 3.949 2.884 2.853 5.607 3.540 14.096 4.060 2.414 6.242 4.559 3.049 2.823 4.307 3.195 2.433 4.125 3.172 3.735 14.388 6.343 3.007 2.529 2.808 2.182 4.467 3.473 14.168 4.724 6.686 5.232 2.624 14.87 993.236 4.501 3.567 9.277 3.293 3.291 4.408 9.892 3.964 3.717 3.665 3.053 2.16 899.899 4.073 2.416 5.387 2.221 4.327 2.826 4.100 3.024 10.58 921.885 14.419 14.817 2.026 3.298 39.63 984.357 2.044 3.765 4.969 2.341 3.954 3.Rev 2/26/14 F-Table for 2.436 3.265 4.061 1.121 3.048 8.222 2.5% points df num 2.364 7.986 2.568 2.22 956.922 5.523 4.819 3.588 3.575 2.197 3.79 799.925 3.329 5.783 3.757 5.116 9.515 3.978 5.102 3.215 3.750 5.307 2.707 2.753 2.152 5.631 2.937 6.66 963.902 2.227 3.529 4.443 16.653 2.512 3.735 2.253 2.090 3.068 1.599 5.767 3.201 4.456 5.312 3.483 3.786 9.506 39.250 3.590 2.074 6.782 2.50 864.011 3.605 7.950 3.695 4.147 3.839 2.101 14.380 3.985 2.667 3.411 2.999 3.669 2.833 8.267 3.063 3.560 6.250 3.694 3.689 9.044 15.078 4.295 3.105 3.945 2.995 4.300 2.507 2.260 6.463 3.756 2.880 2.526 3.571 7.153 4.209 6.248 39.687 2.115 6.200 6.820 5.498 2.547 2.857 4.242 4.395 2.5% F 1 2 3 4 5 6 7 8 9 10 15 20 647.988 5.905 6.788 2.119 4.331 39.374 3.475 3.515 2.026 2.627 2.668 2.061 3.218 10.197 6.934 2.128 3.415 3.874 2.114 8.053 4.167 12.286 5.817 4.440 3.379 3.083 3.542 6.227 5.330 3.199 3.077 4.382 3.769 3.731 2.657 6.474 4.802 2.945 1.285 4.608 3.786 5.613 2.617 2.347 4.209 3.867 2.270 2.389 2.718 4.750 3.364 2.891 3.388 2.681 5.903 3.723 2.576 3.275 4.101 3.922 2.904 2.522 3.28 968.670 3.980 6.319 4.554 6.464 2.388 3.424 5.855 3.373 39.568 5.177 3.871 5.285 3.129 3.903 2.157 2.995 2.774 2.329 3.387 39.798 2.320 4.253 14.182 2.805 3.764 6.965 4.659 5.438 3.156 3.298 6.592 2.007 8.746 2.

359 9 10.802 2.510 4.013 5.815 3.699 3.174 2.057 4.691 3.505 15 8.6 5859.299 3.166 99.635 4.029 5.179 4.994 5.304 3.746 8.675 4.010 3.092 3.579 4.672 10.000 99.979 2.978 2.451 7.333 99.457 28.939 3.051 9.088 2.096 5.162 18 8.849 4.818 3.030 2.546 14.895 3.719 4.256 3.155 8 11.045 3.710 3.664 4.744 4.390 4.522 2.858 13 9.198 120 6.285 6.814 5.636 5.538 4.257 4.699 26 7.539 4.313 3.226 3.823 5.274 12.035 100K 6.324 3.074 6.7 2 98.057 5.558 3.780 3.927 5.640 4.515 5.489 27.636 5.314 6.808 10 10.953 5.260 8.148 8.226 5.802 5.664 27 7.850 2.004 3.739 5.412 5.506 3.499 4.388 3.992 6.718 4.363 3.862 4.620 4.8 6157.466 8.939 3.649 3.126 3.925 9.408 2.945 5.456 10.5 5403.202 4.690 4 21.601 4.032 2.659 14.890 3.511 2.522 15.129 2.953 2.077 19 8.042 3.956 2.840 6.6 5763.547 8.632 2.168 2.926 5.503 99.191 6.434 3.668 5.758 3.799 14.656 3.003 20 8.124 2.649 7.780 9.725 3.258 13.815 2.248 4.312 3.849 4.321 2.499 3.060 11.288 3.177 3.456 4.433 99.330 6.559 2.396 7 12.488 4.640 3.316 5.405 11 9.467 5.302 4.119 2.371 6.400 6.568 4.721 5.874 7.200 5.039 1.102 7.116 30.993 6.106 3.473 3.812 3.Rev 2/26/14 F-Table for 1% Percentage points of the F-distribution: upper 1% points df num 1 2 3 4 5 6 1% F 7 8 9 10 15 20 df den 1 4052.431 4.564 3.437 4.938 21 8.977 4.558 4.805 3.828 3.356 99.938 4.112 5.637 4.627 3.568 4.020 5 16.398 3.453 4.605 3.457 3.140 3.587 3.791 3.888 2.631 3.368 3.559 6.549 40 7.388 99.564 5.069 4.417 4.871 3.598 5.783 2.827 23 7.683 6.310 3.149 3.094 2.719 6.318 4.330 3.667 3.931 2.949 3.372 16 8.259 17 8.064 4.140 4.886 4.710 28.421 3.515 5.976 14.296 4.851 4.185 5.967 10.782 3.158 10.185 4.313 3.665 14 8.319 3.878 K (Multiply this value by 1000) 3-7 .841 3.198 18.258 2.911 5.745 10.632 6.369 60 7.022 6.314 5.785 3.500 4.773 4.640 2.278 4.346 3.206 6.237 27.591 7.017 5.198 3.289 10.765 3.264 3.044 7.178 6.218 3.480 3.632 4.352 2.207 14.597 3.889 2.035 4.100 3.2 4999.018 3.374 99.880 22 7.077 4.962 4.153 3.701 5.753 2.457 3.593 3.562 5.030 3.092 4.217 5.781 24 7.227 3.722 9.514 3.817 29.336 4.895 3.780 4.817 4.345 27.874 4.531 6.358 3.893 4.977 15.217 3.229 26.911 27.3 5981.182 3.602 29 7.074 3.700 2.062 2.449 3 34.103 3.801 2.705 3.099 12 9.561 8.259 8.976 7.5 6055.171 3.026 3.010 4.067 2.523 3.682 3.015 3.369 4.823 2.669 4.359 5.787 3.3 5624.770 5.205 4.142 4.198 14.559 7.862 6.677 5.339 3.453 3.632 28 7.388 4.399 99.256 3.792 2.420 4.422 6.855 3.821 4.006 6.556 4.646 7.1 6022.508 3.718 2.000 16.942 4.552 5.409 3.939 3.988 3.613 5.406 3.120 3.351 5.663 2.0 5928.522 3.005 2.3 6208.291 3.017 2.614 4.591 3.251 4.695 4.993 2.694 15.472 2.620 6.292 4.539 3.699 3.386 5.496 3.726 2.553 6 13.927 3.191 4.754 3.881 5.299 99.102 3.847 7.765 4.441 4.246 9.392 10.574 30 7.460 7.528 3.738 25 7.249 99.526 4.672 27.173 3.872 26.192 2.211 2.

350 3.693 6.814 6.820 4.433 4.102 5.0 198.537 3.492 3.798 10.5 24630.891 6.464 9.40 43.614 12.6 24091.683 3.757 5.742 3.609 4.0 24224.188 2.921 3.280 199.7 22499.176 3.00 49.956 4.883 3.865 5.575 10.956 7.345 4.134 6.215 7.060 10.218 10.151 3.730 3.230 9.272 4.059 3.519 3.541 6.270 4.652 5.952 7.274 4.897 199.471 6.073 9.762 4.570 2.445 4.377 3.339 6.250 8.37 44.496 6.939 3.008 2.991 3.847 3.196 3.715 199.303 6.642 3.663 4.680 6.028 5.384 4.085 20.545 6.139 13.635 9.785 4.983 3.885 7.140 3.450 3.855 4.608 5.202 4.880 3.775 3.776 3.678 7.440 6.855 2.806 6.393 4.427 3.721 4.599 3.345 5.380 7.940 11.556 12.850 3.489 6.701 7.778 20.935 4.904 2.5% Percentage points of the F-distribution: upper 0.717 4.418 5.257 5.823 2.613 3.786 8.681 4.950 4.384 4.155 7.916 5.146 9.5% points df num 1 2 3 4 5 6 0.912 8.462 5.013 2.623 4.799 26.402 3.525 5.913 4.475 9.808 2.262 4.803 5.254 4.551 9.340 4.986 3.000 .880 199.717 8.713 3.322 4.438 13.519 199.387 2.838 21.557 3.047 3.694 6.890 2.928 2.734 3.771 3.409 5.539 5.933 2.200 10.253 5.284 9.561 4.587 3.314 14.059 4.276 5.754 6.028 10.976 4.268 5.682 5.427 8.566 8.424 4.600 7.497 4.50 55.318 3.066 5.791 5.391 8.342 9.649 3.961 10.967 13.150 4.091 5.075 4.300 4.830 9.247 4.093 6.142 4.361 5.013 3.603 4.259 4.228 3.050 8.495 8.514 7.085 4.703 3.226 6.384 10.116 3.740 4.302 7.705 2.2 24836.826 3.389 4.686 20.661 6.893 3.779 4.302 5.8 23437.917 10.519 5.580 3.39 43.645 3.7 19999.374 11.730 5.43 43.513 11.156 6.450 3.109 4.582 5.073 3.374 4.071 5.882 9.727 9.544 12.076 4.559 4.818 5.872 6.090 4.541 5.621 199.828 8.502 3.396 6.698 4.177 4.881 6.5 21614.635 16.243 3.730 6.5% F 7 8 9 10 15 20 df den 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 60 120 ∞ 16210.853 4.042 10.062 3.514 7.372 5.373 2.530 4.890 4.456 14.117 2.354 7.134 2.36 44.343 6.174 5.17 47.926 6.1 23714.835 4.186 7.344 3.375 5.638 5.530 12.521 4.882 21.562 5.598 6.460 4.920 3.202 4.482 5.226 11.211 6.360 3.857 4.087 2.406 9.030 3.944 9.222 3.179 4.33 44.392 22.497 5.498 3.467 24.038 3.259 16.622 14.25 46.956 3.607 3.552 31.492 3.006 2.032 5.103 4.049 4.6 23925.257 4.596 8.233 5.968 2.300 3.167 12.45 42.745 199.352 13.922 7.795 5.832 5.236 14.826 12.116 5.750 3.509 3.195 23.091 3-8 199.239 4.944 3.760 3.291 3.031 4.017 4.486 4.417 5.284 18.071 4.618 10.333 22.30 45.903 9.070 3.688 13.141 4.589 7.317 5.299 199.695 3.212 5.522 8.510 8.975 14.434 21.687 3.885 6.998 5.476 6.043 3.187 199.179 7.754 11.155 15.471 5.692 4.081 7.276 4.073 9.521 6.537 5.968 6.986 6.020 3.811 3.968 5.Rev 2/26/14 F-Table for 0.729 4.126 21.598 2.674 4.246 3.544 4.587 3.785 18.805 7.412 3.933 3.404 11.6 23055.847 5.949 3.793 3.548 3.422 6.110 3.659 4.483 3.781 2.812 3.882 3.847 4.472 4.536 4.107 9.772 10.180 8.814 7.285 3.350 199.355 6.

802 5.692 5.145 60 11.622 7.595 5.139 3.253 5.03 148.761 46.555 4.590 19 15.649 27.894 9.874 3.398 12.873 4.991 4.50 999.828 6.834 28.00 999.461 6.933 4.892 8.698 5.541 3.534 5.759 4.922 7.567 7.637 4.938 6.794 26 13.568 4.438 5.103 3.741 6.534 100K 10.42 4 74.754 5.388 5.804 11 19.245 21.703 21.992 17 15.877 9.330 14.578 9.996 48.752 28.688 18.817 4.885 5.984 5.799 6.238 8.590 5.859 3.554 6.36 999.307 4.096 6.856 7.898 10 21.687 13.656 5.772 7.709 6.658 8.321 7.331 5.460 6.757 4.480 3.193 6.805 6.598 29 13.50 141.006 7.17 999.11 137.241 4.128 10.812 11.335 8.802 6.584 5.926 9.696 6.098 7.463 19.078 5.934 14 17.881 5.054 6.689 18.357 6.237 2.30 999.181 37.85 131.447 4.256 6.292 6.379 8.223 5.274 4.772 17.9K 598.982 6.993 4.953 8.589 5.021 5.326 5.714 11.723 27 13.959 2.283 10.058 23 14.380 7.812 5.109 4.451 6.231 5.25 127.054 4.617 4.829 14.1% Percentage points of the F-distribution: upper 0.643 12.047 8.355 6.128 4.920 3.412 3.781 4.977 5.587 9.4K 585.190 4.973 7.797 4.961 24 14.743 3.559 17.562 6.019 5.100 5 47.251 6.710 7.411 17.763 5.416 4.10 134.406 5.890 4.179 4.186 5.971 9.471 6.044 3.272 6.019 7.848 5.393 4.998 4.1% points df num 1 2 3 4 5 6 7 8 9 10 15 20 df den 1 405.493 40 12.272 6.6K 615.633 8.022 6.198 16.58 132.86 129.8K 620.40 999.866 4.206 15.405 13 17.391 8.832 4.917 25.415 18.019 14.932 8 25.845 5.609 8.191 5.001 7.731 4.485 12.143 11.804 3.107 9.827 120 11.713 4.683 7.704 4.266 K (Multiply entries by 1000 in first row of F values.253 7.222 4.43 999.494 15.326 4.121 6.175 5.712 50.085 29.956 8.355 8.815 12.321 5.593 5.658 28 13.379 3.487 7.803 20.622 6.33 999.040 14.922 7.695 4.239 5. and last value for df) 3-9 .0K 540.195 5.911 25.148 4.223 7.753 3.480 9 22.157 8.392 13.924 20.3K9605.349 3.962 5.086 3.266 3.62 129.535 5.248 16 16.436 51.908 5.075 4.905 12.944 7.517 9.849 7.077 6.5K 576.4K 562.280 7.092 6.459 6.163 27.024 3.265 6.37 126.085 4.858 12.775 18 15.754 8.1K 602.587 11.440 5.560 11.562 4.37 999.204 8.873 25 13.722 10.612 7.059 3.729 8.380 9.070 4.783 2.122 33.553 11.083 13.796 6.739 9.207 4.814 6.292 3.462 5.493 5.767 3.808 6.581 4.116 7.372 4.974 10.655 8.167 22 14.125 5.507 27.308 5.436 7.308 4.687 3.046 11.379 10.558 5.346 9.986 3.489 7.636 4.246 56.498 8.081 5.25 999.206 6.120 10.244 26.557 15 16.525 49.2K 500.561 10.634 14.45 3 167.829 4.947 4.227 3.480 7.030 19.098 2.437 4.857 16.557 5.521 15.028 9.39 999.906 4.390 8.726 5.550 5.290 21 14.543 30 13.395 6 35.758 5.129 7.318 5.171 5.638 4.368 10.513 2.313 10.390 4.116 7.613 9.865 3.081 10.727 7.931 7.902 12.078 2.475 3.552 3.9K 592.804 9.768 6.053 46.436 4.073 8.239 3.422 4.505 4.947 6.404 5.773 7.475 48.841 10.324 12.354 7.137 61.658 48.122 4.481 9.235 4.381 5.469 7.209 9.195 9.202 31.583 6.177 53.9K 2 998.000 23.387 13.008 12 18.819 9.946 4.767 11.293 8.730 4.120 7 29.698 10.400 3.779 9.58 130.339 9.649 5.030 18.430 20 14.339 7.Rev 2/26/14 F-Table for 0.669 6.540 10.

244 2.683 1.827 3.093 2.118 3.092 3.307 2.195 1.443 1.062 2.501 2.846 3.322 1.294 2.671 1.334 1.340 1.419 3.663 2.261 2.424 2.751 3.936 1.218 1.138 2.318 2.330 4.187 3.309 1.925 1.355 1.826 2.667 3.593 4.498 2.805 1.95 7.016 3.865 1.301 3.624 1.216 2.484 3.150 2.538 2.809 3.794 1.169 1.037 2.531 5.006 1. γ = 0.223 3.Rev 2/26/14 Factors for Two-Sided Tolerance Limits for Normal Distributions Factors K2 such that the probability is γ (gamma) that at least a proportion P of the distribution will be included between Y ± K 2s .238 2.948 2.946 3.185 2.568 1.576 P = 0.834 1.166 2.802 4.877 1.183 3.992 1.291 .148 3.593 2.793 3.998 2.757 2.355 2.036 2.431 3.75 4.739 3.979 3.971 2.059 4.846 2.409 1.030 3.844 5.760 1.883 1.117 4.999 11.599 2.588 3.368 3.048 3.889 2.825 1.344 3.521 3.778 3.960 3-10 P = 0.904 2.178 2.704 1.525 1.645 P = 0.033 3.456 2.99 9.911 2.229 2.301 1.609 3.537 2.347 1.871 1.395 1.818 1.375 2.231 1.252 2.337 2.088 2.617 2.728 3.283 1.236 2.636 2.305 1.657 5.920 6.313 1.809 2.262 1.431 4.493 2.271 1.471 4.372 1.709 1.035 1.756 2.429 2.742 1.321 2.328 3.75 n 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 35 40 45 50 75 100 200 500 1000 ∞ P = 0.779 3.169 4.241 4.917 3.465 1.363 1.708 3.317 1.906 1.491 3.915 1.727 2.718 3.635 3.150 P = 0.013 2.962 1.398 2.938 2.444 4.764 3.977 1.860 1.984 2.013 1.948 1.929 2.210 2.328 1.871 2.867 3.297 1.110 4.069 3.177 1.892 2.255 1.891 1.891 3.075 2.383 1.492 1.855 1.677 2.400 3.425 1.90 6.131 2.994 1.414 4.271 3.282 2.271 2.920 2.098 2.898 1.611 3.959 2.222 2.

291 ∞ 3-11 .172 2.264 4.463 2.135 2.943 6.278 5.75 P = 0.999 11.713 3.724 2.805 3.423 2.723 4.089 2.648 2.264 21 1.90 P = 0.615 14 1.717 2.151 25 1.275 1.535 3.562 4.99 P = 0.452 4.054 2.131 3.514 4.203 23 1.822 2.298 1.339 19 1.188 6 2.340 4.306 3.170 4.960 2.506 2.065 2.482 2.743 3.643 3.447 3.520 2.001 3.404 2.232 22 1.594 2.90 n P = 0.764 2.489 3.300 20 1.854 3.506 3.974 11.407 15.098 5.676 3.572 2.251 4.471 2.167 30.576 3.626 3.959 2.974 35 1.046 10 1.588 3.246 2.199 4.935 2.775 2.521 5.044 2.919 8.543 3.390 1000 1. γ = 0.762 3.176 24 1.800 24.413 3.292 4.085 28 1.863 3.132 5.Rev 2/26/14 Factors for Two-Sided Tolerance Limits for Normal Distributions Factors K2 such that the probability is γ (gamma) that at least a proportion P of the distribution will be included between Y ± K 2s .750 7 1.219 2.066 29 1.106 27 1.435 3.454 3.417 2.839 2.150 1.284 3.152 2.917 40 1.424 3.552 2.034 2.077 2.125 4.433 4.220 9 1.906 11 1.278 2.833 50 1.382 18 1.227 2 4.232 4.493 2.025 2.529 200 1.340 1.870 6.314 2.315 4.184 4.849 4.034 2.460 3.211 2.234 1.921 2.434 500 1.695 2.906 3.049 30 1.916 2.932 4.682 4.756 3.545 15 1.645 1.564 3.430 17 1.683 2.959 5.196 3.354 1.689 3.112 3.847 6.484 16 1.334 3.270 4.535 2.127 26 1.494 4.654 3.201 1.437 2.018 3.444 2.856 2.902 3.355 2.988 2.172 2.758 4.370 1.646 100 1.030 3.524 3.102 2.149 4 2.446 8 1.368 4.368 3.423 6.879 5 2.215 4.792 12 1.614 3.019 2.399 4.95 P = 0.166 4.152 5.697 13 1.871 45 1.390 1.066 3.453 2.619 2.194 2.474 3.430 2.440 8.471 4.462 2.712 75 1.046 2.933 3.185 1.618 4.118 2.103 2.978 18.309 3 2.

899 9 1.993 50 1.775 7.579 3.634 8.798 2.614 20 1.858 2.012 3.380 9.379 4.996 2.615 4.264 2.567 21 1.676 7 2.655 2.150 5.576 3.484 23 1.178 2.079 6.878 4.418 1000 1.310 2.90 P = 0.215 1.810 2.002 4.557 2.858 3.259 4.311 1.732 4.95 n P = 0.272 4.368 4.651 3.674 48.816 3.Rev 2/26/14 Factors for Two-Sided Tolerance Limits for Normal Distributions Factors K2 such that the probability is γ (gamma) that at least a proportion P of the distribution will be included between Y ± K 2s .532 4.415 5 2.735 2.433 5.299 10.587 3.165 4.770 2.480 2.955 5.291 ∞ 3-12 .604 3.752 3.278 30 1.301 29 1.754 4.337 6 2.036 2.916 12.447 24 1.369 6.021 2.545 2.445 3.709 2.197 3.819 3.457 4.475 500 1.396 1.523 2.534 2.737 2.712 4.514 2.490 3.042 45 1.645 1.052 2.779 5.697 3.435 2.649 10 1.595 3.208 2.543 4.723 3.960 2.370 8. γ = 0.835 75 1.379 3.564 3.737 3.140 2.865 16 1.573 2 5.143 2.549 3.597 200 1.366 2.839 3.616 2.462 2.577 4.341 1.954 3.400 2.326 28 1.430 60.891 6.705 2.150 1.179 35 1.388 4.244 2.019 37.676 3.967 3.158 13 1.916 2.408 3.337 2.523 22 1.382 26 1.409 4.126 3.361 3.505 2.369 4.078 2.193 2.529 3.721 3.999 22.208 3 3.414 5.903 3.748 100 1.667 19 1.044 5.725 18 1.599 2.007 5.702 4.277 5.103 40 1.045 14 1.248 6.225 2.631 3.949 15 1.285 3.483 4.95 P = 0.874 2.922 8.987 2.791 17 1.233 2.081 4.414 2.258 1.655 3.570 2.673 3.226 8 2.631 5.162 4.656 4.275 5.164 2.291 12 1.002 3.432 4.497 2.512 4.350 4.584 2.75 P = 0.353 27 1.502 4 3.99 P = 0.812 4.413 25 1.070 2.437 2.452 11 1.858 32.934 3.917 2.136 3.212 4.286 2.612 3.861 16.635 2.090 2.152 2.195 1.679 2.784 3.

936 6.905 4.643 6.378 18.079 6.764 2.044 5.527 18.872 3.221 4.130 3.054 2 13.805 29 1.383 1.75 P = 0.117 2.808 2.150 1.600 9 2.841 3.732 16 1.865 2.240 13 2.659 3.469 2.977 2.399 19 1.777 2.187 9.279 4.150 14.120 3.345 8.234 8 2.300 303.243 1.954 100 1.068 2.444 4.860 2.616 3 6.098 23 1.197 4.945 3.Rev 2/26/14 Factors for Two-Sided Tolerance Limits for Normal Distributions Factors K2 such that the probability is γ (gamma) that at least a proportion P of the distribution will be included between Y ± K 2s .733 4.385 4.260 13.727 2.763 4.608 4.009 2.055 36.926 2.914 3.904 4.147 4.363 160.681 2.966 7.845 28 1.703 3.90 P = 0.822 4.736 2.551 3.548 6 3.100 5.234 21 1.711 2.695 2.508 3.960 2.060 2.492 5.876 15 2.488 7.99 P = 0.539 2.129 10 2.993 5.433 3.308 6.888 27 1.745 2.214 1.785 2.161 5.378 3.753 3.584 3.043 14 2.200 2.611 35 1.162 2.947 5.794 4.731 200 1.096 3.230 5.029 3.039 24 1.605 5.870 5.345 4.307 5.045 5.555 500 1.576 3.397 4.428 2.613 2.142 7 2.193 188.743 5.783 3.613 5.594 7.576 3.930 22.078 4.306 2.004 3.99 n P = 0.677 3.163 22 1.865 4.550 5.401 29.250 3.833 2.015 5 3.571 2.935 26 1.385 2.748 3.383 4 4.612 7.965 2.620 3.274 3.399 45 1.855 10.355 3.424 2.999 114.304 1.337 6.393 5.265 5.301 10.512 2.95 P = 0.494 2.668 2.491 242.941 3.421 4.611 4.247 2.893 6.718 3.291 ∞ 3-13 .607 17 1.864 3.233 4.828 4.921 3.985 25 1.766 11 2.614 9.888 3. γ = 0.582 4.312 20 1.808 3.477 12 2.472 1000 1.222 2.621 3.737 6.768 30 1.891 2.121 4.493 40 1.040 3.468 8.507 4.518 4.042 2.522 3.323 50 1.645 1.168 4.677 3.972 3.497 18 1.446 2.398 11.727 4.190 3.084 75 1.404 2.

396 5.977 3.838 1.595 18 0.90 P=0.458 1.825 1.834 2.841 1.976 2.645 2.492 1.726 4.859 1.722 13 0.895 2.461 29 0.927 3.932 2.675 3.577 1.521 1.088 1.788 1.026 2.920 1.421 35 0.947 1.932 1.464 2.95 P=0.723 3.901 2.273 6 1.681 3.507 5 1.633 3.854 1.119 7 1.803 1.042 4.697 3.614 3.048 2.606 2.565 1.520 23 0.710 3.911 4 1.445 1.497 25 0.805 3 1.454 30 0.775 3.690 14 0.883 2.885 3.869 2.811 2.546 21 0.620 3.962 2.878 2.244 4.Rev 2/26/14 Factors for One-Sided Tolerance Limits for Normal Distributions Factors K1 such that the probability is γ (gamma) that at least a proportion P of the distribution will be less than Y + K 1s or greater than Y − K 1s .75 n P=0.554 1.859 2.552 3.795 1.250 3.614 17 0.073 2.991 2.845 1.924 9 0.487 1.640 3.952 2.528 1.915 2.536 1.75 P=0.813 1.661 15 0.822 3.665 3.043 1.375 45 0.126 4.141 2.104 2.636 16 0.624 2.909 1.999 1.873 2.804 11 0.883 1.188 3.985 1.010 1.463 3.090 ∞ 3-14 .941 2.889 2.849 1.469 28 0.282 1.828 1.790 2.152 1.577 19 0.134 2.591 2.008 8 0.425 1.891 1.851 3.645 2.336 3.99 P=0.487 26 0.923 2.876 1.963 2.545 1.508 1. γ = 0.514 1.502 1.701 2.007 2.858 10 0.870 1.864 1.560 20 0.797 3.475 1.760 12 0.849 2.756 3.674 1.588 3.648 3.152 4.326 3.626 3.538 3.685 3.255 2.358 50 0.831 1.899 1.501 3.821 2.478 1.421 4.964 1.671 2.434 1.908 2.656 3.568 3.478 27 0.739 3.834 1.483 1.508 24 0.497 1.532 22 0.395 40 0.740 2.

985 1.503 3.800 2.942 1.272 3.282 1.967 1.514 11 1.99 P=0.556 6 1.679 40 0.674 1.979 1.077 4.737 2.750 2.443 4.783 4.819 2.041 2.826 28 0.842 2.159 2.986 2.363 3.843 27 0.598 2.243 5.299 3.164 16 1.895 2.052 4.309 4.577 1.894 1.972 5.691 2.400 4.402 3.007 3.371 4.139 1.742 3.833 3.078 18 1.965 2.119 1.438 7.326 3.215 15 1.435 2.969 3.080 2.129 4 1.540 2.016 1.955 8 1.066 2.249 3.111 5 1.884 3.765 2.099 2.273 14 1.90 n P=0.867 2.132 2.896 3.603 4.712 2.109 2.028 3.340 9.730 35 0.120 2.928 2.219 2.75 P=0.258 5.937 3.973 1.042 19 1.010 2.650 3.333 2.999 2.952 22 1.793 3.894 3.188 1.698 2.782 2.761 3.448 3.952 3.922 3.638 45 0.90 P=0.666 6.771 9 1.702 2.101 1.025 1.95 P=0.119 17 1.903 24 1.035 1.494 3.966 2.645 2.329 3.137 4.657 2.302 2.651 3 1.624 2.992 1.208 3.058 1.735 3.227 3.568 3.641 4.Rev 2/26/14 Factors for One-Sided Tolerance Limits for Normal Distributions Factors K1 such that the probability is γ (gamma) that at least a proportion P of the distribution will be less than Y + K 1s or greater than Y − K 1s γ = 0.257 2.665 2.046 1.219 2.957 5.092 4.794 30 0.257 4.754 3.090 ∞ 3-15 .162 1.190 3.810 29 0.559 1.011 2.724 2.007 1.188 3.341 13 1.927 23 1.882 25 0.212 4.105 4.174 3.629 10 1.000 1.909 3.133 2.009 20 1.532 4.311 7.360 2.682 2.979 21 1.145 2.907 1.862 26 0.202 7 1.604 50 0.071 1.085 1.987 3.673 2.420 12 1.172 4.089 2.923 1.972 3.

407 4.143 5.166 1.187 4.941 3.618 2.092 2.612 6 1.201 1.886 2.414 4.328 3.582 3.090 ∞ 3-16 .155 7.260 3.058 1.852 5.732 2.002 2.033 2.898 3.566 3.75 P=0.806 6.318 20 1.354 5.857 3 2.042 9.999 3.144 7.277 21 1.136 4.659 4.068 2.142 25 1.066 28 1.585 4.210 2.454 3.080 4.788 2.204 23 1.295 4.978 1.089 27 1.328 2.671 3.900 12 1.155 2.614 3.203 10 1.999 1.371 3.787 13 1.349 3.025 1.239 22 1.355 2.370 4.862 3.411 2.95 P=0.275 2.364 19 1.126 2.098 4.646 2.162 5.960 1.815 3.423 3.995 3.532 2.911 3.465 2.062 6.220 3.838 2.618 4.674 1.243 2.036 11 1.Rev 2/26/14 Factors for One-Sided Tolerance Limits for Normal Distributions Factors K1 such that the probability is γ (gamma) that at least a proportion P of the distribution will be less than Y + K 1s or greater than Y − K 1s γ = 0.263 4.366 2.811 2.063 7 1.399 4.811 45 0.732 2.125 1.747 4.109 2.103 1.114 1.642 6.866 40 0.066 1.232 3.502 5 1.553 13.296 2.415 18 1.268 2.309 3.934 35 0.853 2.926 2.824 2.206 4.777 2.214 4 2.535 16 1.90 P=0.220 2.645 2.006 3.075 1.905 2.150 3.736 3.282 1.292 3.043 29 1.697 2.064 4.95 n P=0.326 3.233 4.869 2.471 17 1.755 3.172 24 1.158 4.083 1.167 2.974 2.183 1.065 2.656 10.022 30 1.708 5.895 3.203 5.152 1.181 4.413 9 1.690 14 1.115 26 1.799 2.453 3.246 3.981 5.949 2.138 1.117 4.486 3.275 3.607 15 1.99 P=0.766 50 0.331 4.741 7.396 3.688 8 1.669 2.524 3.031 4.464 4.520 4.093 1.

326 3.508 30 1.180 2.389 7.450 2.502 4.504 15 1.090 ∞ 3-17 .662 3.808 23 1.189 4.273 2.411 5.472 5.028 4.249 4.001 20 1.288 11 1.410 4.706 25 1.677 3.Rev 2/26/14 Factors for One-Sided Tolerance Limits for Normal Distributions Factors K1 such that the probability is γ (gamma) that at least a proportion P of the distribution will be less than Y + K 1s or greater than Y − K 1s γ = 0.606 10 1.738 5.755 24 1.514 2.209 2.355 2.857 2.581 3.821 2.399 2.085 2.312 3.405 2.168 45 1.030 2.633 3.564 35 1.276 2.269 3.335 9.080 19 1.674 1.867 22 1.377 16 1.777 4.376 2.154 1.812 7.957 2.491 3.473 4.771 2.154 2.827 13 1.618 27 1.727 4.336 2.346 7 2.074 6.497 4.853 2.902 2.558 3.102 4.047 2.854 3.285 5.337 5.037 5.593 3.035 12 1.848 4.412 8.606 3.066 2.447 4.564 8 2.096 50 0.536 3.633 6.892 5.832 5.777 3.123 5.905 3.703 2.522 3.290 4.601 4.319 2.808 3.364 3.640 4.556 4.579 28 1.552 2.681 4.99 n P=0.459 3.972 5.121 1.241 3.222 5.255 40 1.260 2.430 3.303 2.314 2.247 2.167 18 1.729 3.652 14 1.963 4.241 2.287 2.282 1.533 4.954 3.960 5.129 2.334 4.550 6 2.406 7.180 4.931 21 1.728 6.195 1.542 29 1.999 2.083 3.106 2.595 2.516 3.094 1.048 3.859 4.75 P=0.829 6.015 9 1.645 2.423 2.694 3.99 P=0.357 2.566 4.898 3.253 3.660 26 1.645 2.90 P=0.766 3.265 17 1.481 2.125 4.95 P=0.

2 11.3 2.6 3.14 6.13 2.11 2.7 3.96 37.1 500 59.7 3.12 5.1 9.75 0. 3-18 0.21 8.2 80 8.95 0.9 4.1 95 11.3 2.30 14.2 12.1 45.5 2.13 5.1 100 11.2 1.2 150 17.5 2.2 60 7.32 12.18 9.27 13.4 2.15 3.1 1.2 1.11 2.6 1.1 130 15.1 10.32 16.47 23.20 4.1 9.17 6.3 2.9 1.75 n\P 0.37 18.83 33.18 3.16 8.7 1.1 110 12.2 14.4 21.3 When the values of r and s are not equal they are interchangeable.7 4. 120 47.14 5.8 3.5 2. s) such that we may assert with at least γ (gamma) confidence that 100P percent of the population lies between the rth smallest and the sth largest of a random sample of n.4 2.2 68.25 12. γ = 0.20 3.7 3.2 1.108 42.2 800 96.17 6.1 7.1 120 14.1 140 16.12 4.99 0.6 3.11 4.3 92. 6 2.42 21.44 21.18 3.8 3.1 90 10.8 2.71 28.68 26.10 4.5 2.5 400 47.10 4.5 3.11 4.2 1.2 1.92 35.90 5.7 3.10 3.4 2.6 2.9 3.2 70 8.56 21.90 50 5.8 4.4 117.35 13.3 1.1 8. 5 2.1 8.3 18.3 1.1 6.44 16.4 2.5 2.13 5.6 3.1 5.116 44.13 6.3 2.47 18.1 7.90 0.8 3.19 7.22 4.15 6.15 5.2 1.1 85 10.2 13.95 0.11 5.3 300 35.1 700 84.59 23.34 16.99 .1 600 72.3 2.14 2.2 900 108.1 33.1 6.Rev 2/26/14 Distribution-Free Two-Sided Tolerance Limits Values (r.3 104.8 3.22 11.2 200 23.3 1.6 65 γ = 0.75 0.1 7.1 57.4 1.2 1000 121.20 10.104 40.2 75 8.7 2.39 19.23 9.3 16.80 30.1 55 6.16 3.2 80.6 2.2 170 20.1 2.

4 2.13 4.6 2.3 1.26 11.7 2.2 6.2 110.22 9.95 0.54 20.1 80 7.6 3.3 1.5 500 55.43 15.11 5.16 6. 3-19 0.1 12.7 2.2 300 32.7 2. 114 43.74 26.1 100 9.11 4.2 98.1 5.1 3.42 20.97 35.13 4.90 33.10 3. 109 40.29 10.13 1.11 1.3 1.1 800 90.1 5.9 3.40 14.1 4.2 15.37 18.6 2.14 5.31 11.33 15.8 3.5 2.1 75.6 2.75 0.2 110 10.11 3.2 120 11.2 7.17 2.5 1.1 52.1 70 6.2 130 13.99 n\P 0.1 When the values of r and s are not equal they are interchangeable.95 0.15 5.63 22.3 60 5.4 65 6.18 6.7 600 67.66 24.1 85 8.19 3.19 9. γ = 0.5 2.4 2.1 170 17.3 1.10 3.8 3.5 1.8 1.3 400 43.5 1.1 63.1 900 102.1 5.1 75 7.2 18.10 1.1 140 14.3 1.15 2.1 1000 114. 3 55 5.4 2.12 4.1 7.1 8.13 6.99 . 4 1.5 1.9 4.20 7.17 7.102 38.78 29.2 1.8 2.4 1.9 700 78.4 29.6 40.90 0.1 90 8.2 86.90 50 4.17 2.13 2.99 0.75 0.15 7.6 2.1 13.1 9.5 2. s) such that we may assert with at least γ (gamma) confidence that 100P percent of the population lies between the rth smallest and the sth largest of a random sample of n.39 18.7 2.28 13.Rev 2/26/14 Distribution-Free Two-Sided Tolerance Limits Values (r.2 6.1 150 15.15 2.1 200 20.95 γ = 0.1 11.35 15. 4 1.4 2.1 4.51 18.24 11.1 95 9.2 1.9 3.30 13.86 31.1 10.

95 0.95 0.95 γ = 0.75 0.90 0.75 0.95 0.90 0.99 0.Rev 2/26/14 Distribution-Free One-Sided Tolerance Limits Values (m) such that with at least γ (gamma) confidence that 100P percent of the population lies below the mth largest (or above the mth smallest) of a random sample of n.75 γ = 0.90 0.99 n\P 0.75 0.99 0.90 0.99 50 10 3 1 9 2 1 8 2 6 1 55 12 4 2 10 3 1 9 2 7 1 60 13 4 2 11 3 1 10 2 1 8 1 65 14 5 2 12 4 1 11 3 1 9 2 70 15 5 2 13 4 1 12 3 1 10 2 75 16 6 2 14 4 1 13 3 1 10 2 80 17 6 3 15 5 2 14 4 1 11 2 85 19 7 2 16 5 2 15 4 1 12 3 90 20 7 2 17 5 2 16 5 1 13 3 1 95 21 7 2 18 6 2 17 5 2 14 3 1 100 22 8 2 20 6 2 18 5 2 15 4 1 110 24 9 4 22 7 3 20 6 2 17 4 1 120 27 10 4 24 8 3 22 7 2 19 5 1 130 29 11 5 26 9 3 25 8 3 21 6 2 140 31 12 5 1 28 10 4 27 8 3 23 6 2 150 34 12 6 1 31 10 4 29 9 3 26 7 2 170 39 14 7 1 35 12 5 33 11 4 30 9 3 200 46 17 8 1 42 15 6 40 13 5 36 11 4 300 70 26 12 2 65 23 10 1 63 22 9 1 58 19 7 400 94 36 17 3 89 32 15 2 86 30 13 1 80 27 11 500 118 45 22 3 113 41 19 2 109 39 17 2 103 35 14 1 600 143 55 26 4 136 51 23 3 133 48 21 2 126 44 18 1 700 167 65 31 5 160 60 28 4 156 57 26 3 149 52 22 2 800 192 74 36 6 184 69 32 5 180 66 30 4 172 61 26 2 900 216 84 41 7 208 79 37 5 204 75 35 4 195 70 30 3 1000 241 94 45 8 233 88 41 6 228 85 39 5 219 79 35 3 3-20 .75 0.90 γ = 0. γ = 0.95 0.99 0.