DOE Glossary: Difference between revisions

From ReliaWiki
Jump to navigation Jump to search
(Created page with 'Glossary Alias Two or more effects are said to be aliased in an experiment if these effects cannot be distinguished from each other. This happens when the columns of the design…')
 
No edit summary
Line 1: Line 1:
Glossary  
=Glossary=
Alias
 
'''Alias'''


Two or more effects are said to be aliased in an experiment if these effects cannot be distinguished from each other. This happens when the columns of the design matrix corresponding to these effects are identical. As a result, the aliased effects are estimated by the same linear combination of observations instead of each effect being estimated by a unique combination.
Two or more effects are said to be aliased in an experiment if these effects cannot be distinguished from each other. This happens when the columns of the design matrix corresponding to these effects are identical. As a result, the aliased effects are estimated by the same linear combination of observations instead of each effect being estimated by a unique combination.
Line 6: Line 7:
   
   


ANOVA
'''ANOVA'''


ANOVA is the acronym for Analysis of Variance. It refers to the procedure of splitting the variability of a data set to conduct various significance tests.
ANOVA is the acronym for Analysis of Variance. It refers to the procedure of splitting the variability of a data set to conduct various significance tests.
Line 12: Line 13:
   
   


ANOVA Model
'''ANOVA Model'''


The regression model where all factors are treated as qualitative factors. ANOVA models are used in the analysis of experiments to identify significant factor(s) by investigating each level of the factor(s) individually.
The regression model where all factors are treated as qualitative factors. ANOVA models are used in the analysis of experiments to identify significant factor(s) by investigating each level of the factor(s) individually.
Line 18: Line 19:
   
   


Balanced Design
'''Balanced Design'''


An experiment in which equal number of observations are taken for each treatment.
An experiment in which equal number of observations are taken for each treatment.
Line 24: Line 25:
   
   


Blocking
'''Blocking'''


Separation of experiment runs based on the levels of a nuisance factor. Blocking is used to deal with known nuisance factors. You should block what you can and randomize what you cannot. See also Nuisance Factors, Randomization.
Separation of experiment runs based on the levels of a nuisance factor. Blocking is used to deal with known nuisance factors. You should block what you can and randomize what you cannot. See also Nuisance Factors, Randomization.
Line 30: Line 31:
   
   


Center Point
'''Center Point'''


The experiment run that corresponds to the mid-level of all the factor ranges.
The experiment run that corresponds to the mid-level of all the factor ranges.
Line 36: Line 37:
   
   


Coded Values
'''Coded Values'''


The factor values that are such that the upper limit of the investigated range of the factor becomes  and the lower limit becomes . Using coded values makes the experiments with all factors at two levels orthogonal.
The factor values that are such that the upper limit of the investigated range of the factor becomes  and the lower limit becomes . Using coded values makes the experiments with all factors at two levels orthogonal.
Line 42: Line 43:
   
   


Confidence Interval
'''Confidence Interval'''


A closed interval where a certain percentage of the population is likely to lie. For example, a 90% confidence interval with a lower limit of  and an upper limit of  implies that 90% of the population lies between the values of  and .
A closed interval where a certain percentage of the population is likely to lie. For example, a 90% confidence interval with a lower limit of  and an upper limit of  implies that 90% of the population lies between the values of  and .
Line 48: Line 49:
   
   


Confounding
'''Confounding'''


Confounding occurs in a design when certain effects cannot be distinguished from the block effect. This happens when full factorial designs are run using incomplete blocks. In such designs the same linear combination of observations estimates the block effect and the confounded effects. See also Incomplete Blocks.
Confounding occurs in a design when certain effects cannot be distinguished from the block effect. This happens when full factorial designs are run using incomplete blocks. In such designs the same linear combination of observations estimates the block effect and the confounded effects. See also Incomplete Blocks.
Line 54: Line 55:
   
   


Contrast
'''Contrast'''


Any linear combination of two or more factor level means such that the coefficients in the combination add up to zero. The difference between the means at any two levels of a factor is an example of a contrast.
Any linear combination of two or more factor level means such that the coefficients in the combination add up to zero. The difference between the means at any two levels of a factor is an example of a contrast.
Line 60: Line 61:
   
   


Control Factors
'''Control Factors'''


The factors affecting the response that are easily manipulated and set by the operator. See also Noise Factors.
The factors affecting the response that are easily manipulated and set by the operator. See also Noise Factors.
Line 66: Line 67:
   
   


Cross Array Design
'''Cross Array Design'''


The experiment design in which every treatment of the inner array is replicated for each run of the outer array. See also Inner Array, Outer Array.
The experiment design in which every treatment of the inner array is replicated for each run of the outer array. See also Inner Array, Outer Array.
Line 72: Line 73:
   
   


Curvature Test
'''Curvature Test'''


The test that investigates if the relation between the response and the factors is linear by using center points. See also Center Point.
The test that investigates if the relation between the response and the factors is linear by using center points. See also Center Point.
Line 78: Line 79:
   
   


Defining Relation
'''Defining Relation'''


For two level fractional factorial experiments, the equation that is used to obtain the fraction from the full factorial experiment. The equation shows which of the columns of the design matrix in the fraction are identical to the first column. For example, the defining relation  can be used to obtain a half-fraction of the two level full factorial experiment with three factors ,  and . The effect(s) used in the equation is called the generator or word.
For two level fractional factorial experiments, the equation that is used to obtain the fraction from the full factorial experiment. The equation shows which of the columns of the design matrix in the fraction are identical to the first column. For example, the defining relation  can be used to obtain a half-fraction of the two level full factorial experiment with three factors ,  and . The effect(s) used in the equation is called the generator or word.
Line 84: Line 85:
   
   


Degrees of Freedom
'''Degrees of Freedom'''


The number of independent observations made in excess of the unknowns.
The number of independent observations made in excess of the unknowns.
Line 90: Line 91:
   
   


Design Matrix
'''Design Matrix'''


The matrix whose columns correspond to the levels of the variables (and their interactions) at which observations are recorded.
The matrix whose columns correspond to the levels of the variables (and their interactions) at which observations are recorded.
Line 96: Line 97:
   
   


Design Resolution
'''Design Resolution'''


The number of factors in the smallest word in a defining relation. Design resolution indicates the degree of aliasing in a fractional factorial design. See also Defining Relation, Word.
The number of factors in the smallest word in a defining relation. Design resolution indicates the degree of aliasing in a fractional factorial design. See also Defining Relation, Word.
Line 102: Line 103:
   
   


Error
'''Error'''


The natural variations that occur in a process, even when all the factors are maintained at the same level. See also Residual.
The natural variations that occur in a process, even when all the factors are maintained at the same level. See also Residual.
Line 108: Line 109:
   
   


Error Sum of Squares
'''Error Sum of Squares'''


The variation in the data not captured by the model. The error sum of squares is also called the residual sum of squares. See also Model Sum of Squares, Total Sum of Squares.
The variation in the data not captured by the model. The error sum of squares is also called the residual sum of squares. See also Model Sum of Squares, Total Sum of Squares.
Line 114: Line 115:
   
   


Extra Sum of Squares
'''Extra Sum of Squares'''


The increase in the model sum of squares when a term is added to the model.
The increase in the model sum of squares when a term is added to the model.
Line 120: Line 121:
   
   


Factorial Experiment
'''Factorial Experiment'''


The experiment in which all combinations of the factor levels are run.
The experiment in which all combinations of the factor levels are run.
Line 126: Line 127:
   
   


Fractional Factorial Experiment
'''Fractional Factorial Experiment'''


The experiment where only a fraction of the combinations of the factor levels are run.
The experiment where only a fraction of the combinations of the factor levels are run.
Line 132: Line 133:
   
   


Factor
'''Factor'''


The entity whose effect on the response is investigated in the experiment.
The entity whose effect on the response is investigated in the experiment.
Line 138: Line 139:
   
   


Fitted Value
'''Fitted Value'''


The estimate of an observation obtained using the model that has been fit to all the observations.
The estimate of an observation obtained using the model that has been fit to all the observations.
Line 144: Line 145:
   
   


Fixed Effects Model
'''Fixed Effects Model'''


The ANOVA model used in the experiments where only a limited number of the factor levels are of interest to the experimenter. See also Random Effects Model.
The ANOVA model used in the experiments where only a limited number of the factor levels are of interest to the experimenter. See also Random Effects Model.
Line 150: Line 151:
   
   


Full Model
'''Full Model'''


The model that includes all the main effects and their interactions. In DOE++, a full model is the model that contains all the effects that are specified by the user. See also Reduced Model.
The model that includes all the main effects and their interactions. In DOE++, a full model is the model that contains all the effects that are specified by the user. See also Reduced Model.
Line 156: Line 157:
   
   


Generator
'''Generator'''


See Word.
See Word.
Line 162: Line 163:
   
   


Hierarchical Model
'''Hierarchical Model'''


In DOE++, a model is said to be hierarchical, if, corresponding to every interaction, the main effects of the related factors are included in the model.
In DOE++, a model is said to be hierarchical, if, corresponding to every interaction, the main effects of the related factors are included in the model.
Line 168: Line 169:
   
   


Incomplete Blocks
'''Incomplete Blocks'''


Blocks that do not contain all the treatments of a factorial experiment.
Blocks that do not contain all the treatments of a factorial experiment.
Line 174: Line 175:
   
   


Inner Array
'''Inner Array'''


The experiment design used to investigate the control factors under Taguchi's philosophy to design a robust system. See also Robust System, Outer Array, Cross Array.
The experiment design used to investigate the control factors under Taguchi's philosophy to design a robust system. See also Robust System, Outer Array, Cross Array.
Line 180: Line 181:
   
   


Interactions
'''Interactions'''


Interaction between factors means that the effect produced by a change in a factor on the response depends on the level of the other factor(s).
Interaction between factors means that the effect produced by a change in a factor on the response depends on the level of the other factor(s).
Line 186: Line 187:
   
   


Lack-of-Fit Sum of Squares
'''Lack-of-Fit Sum of Squares'''


The portion of the error sum of squares that represents variation in the data not captured because of using a reduced model. See also Reduced Model, Pure Error Sum of Squares.
The portion of the error sum of squares that represents variation in the data not captured because of using a reduced model. See also Reduced Model, Pure Error Sum of Squares.
Line 192: Line 193:
   
   


Least Squares Means
'''Least Squares Means'''


The predicted mean response value for a given factor level while the remaining factors in the model are set to the coded value of zero.
The predicted mean response value for a given factor level while the remaining factors in the model are set to the coded value of zero.
Line 198: Line 199:
   
   


Level
'''Level'''


The setting of a factor used in the experiment.
The setting of a factor used in the experiment.
Line 204: Line 205:
   
   


Main Effect
'''Main Effect'''


The change in the response due to a change in the level of a factor.
The change in the response due to a change in the level of a factor.
Line 210: Line 211:
   
   


Mean Square
'''Mean Square'''


The sum of squares divided by the respective degrees of freedom.
The sum of squares divided by the respective degrees of freedom.
Line 216: Line 217:
   
   


Model Sum of Squares
'''Model Sum of Squares'''


The portion of the total variability in the data that is explained by the model. See also Error Sum of Squares, Total Sum of Squares.
The portion of the total variability in the data that is explained by the model. See also Error Sum of Squares, Total Sum of Squares.
Line 222: Line 223:
   
   


Multicollinearity
'''Multicollinearity'''


A model with strong dependencies between the independent variables is said to have multicollinearity.
A model with strong dependencies between the independent variables is said to have multicollinearity.
Line 228: Line 229:
   
   


New Observations
'''New Observations'''


Observations that are not part of the data set used to fit the model.
Observations that are not part of the data set used to fit the model.
Line 234: Line 235:
   
   


Noise Factors
'''Noise Factors'''


Those nuisance factors that vary uncontrollably or naturally and can only be controlled for experimental purposes. For example, ambient temperature, atmospheric pressure and humidity are examples of noise factors.
Those nuisance factors that vary uncontrollably or naturally and can only be controlled for experimental purposes. For example, ambient temperature, atmospheric pressure and humidity are examples of noise factors.
Line 240: Line 241:
   
   


Nuisance Factors
'''Nuisance Factors'''


Factors that have an effect on the response but are not of primary interest to the investigator.
Factors that have an effect on the response but are not of primary interest to the investigator.
Line 246: Line 247:
   
   


Orthogonal Array
'''Orthogonal Array'''


An array in which all the columns are orthogonal to each other. Two columns are said to be orthogonal if the sum of the terms resulting from the product of the columns is zero.
An array in which all the columns are orthogonal to each other. Two columns are said to be orthogonal if the sum of the terms resulting from the product of the columns is zero.
Line 252: Line 253:
   
   


Orthogonal Design
'''Orthogonal Design'''


An experiment design is orthogonal if the corresponding design matrix is such that the sum of the terms resulting from the product of any two columns is zero. In orthogonal designs the analysis of an effect does not depend on what other effects are included in the model.
An experiment design is orthogonal if the corresponding design matrix is such that the sum of the terms resulting from the product of any two columns is zero. In orthogonal designs the analysis of an effect does not depend on what other effects are included in the model.
Line 258: Line 259:
   
   


Outer Array
'''Outer Array'''


The experiment design used to investigate noise factors under Taguchi's philosophy to design a robust system. See also Robust System, Inner Array, Cross Array.
The experiment design used to investigate noise factors under Taguchi's philosophy to design a robust system. See also Robust System, Inner Array, Cross Array.
Line 264: Line 265:
   
   


Partial Sum of Squares
'''Partial Sum of Squares'''


The type of extra sum of squares that is calculated assuming that all terms other than the given term are included in the model. The partial sum of squares is also referred to as the adjusted sum of squares. See also Extra Sum of Squares, Sequential Sum of Squares.
The type of extra sum of squares that is calculated assuming that all terms other than the given term are included in the model. The partial sum of squares is also referred to as the adjusted sum of squares. See also Extra Sum of Squares, Sequential Sum of Squares.
Line 270: Line 271:
   
   


Prediction Interval
'''Prediction Interval'''


The confidence interval on new observations.
The confidence interval on new observations.
Line 276: Line 277:
   
   


Pure Error Sum of Squares
'''Pure Error Sum of Squares'''


The portion of the error sum of squares that represents variation due to replicates. See also Lack-of-Fit Sum of Squares.
The portion of the error sum of squares that represents variation due to replicates. See also Lack-of-Fit Sum of Squares.
Line 282: Line 283:
   
   


Qualitative Factor
'''Qualitative Factor'''


The factor where the levels represent different categories and no numerical ordering is implied. These factors are also called categorical factors.
The factor where the levels represent different categories and no numerical ordering is implied. These factors are also called categorical factors.
Line 288: Line 289:
   
   


Random Effects Model
'''Random Effects Model'''


The ANOVA model used in the experiments where the factor levels to be investigated are randomly selected from a large or infinite population. See also Fixed Effects Model.
The ANOVA model used in the experiments where the factor levels to be investigated are randomly selected from a large or infinite population. See also Fixed Effects Model.
Line 294: Line 295:
   
   


Randomization
'''Randomization'''


Conducting experiment runs in a random order to cancel out the effect of unknown nuisance factors. See also Blocking.
Conducting experiment runs in a random order to cancel out the effect of unknown nuisance factors. See also Blocking.
Line 300: Line 301:
   
   


Randomized Complete Block Design
'''Randomized Complete Block Design'''


An experiment design where each block contains one replicate of the experiment and runs within the block are subjected to randomization.
An experiment design where each block contains one replicate of the experiment and runs within the block are subjected to randomization.
Line 306: Line 307:
   
   


Reduced Model
'''Reduced Model'''


A model that does not contain all the main effects and interactions. In DOE++, a reduced model is the model that does not contain all the effects specified by the user. See also Full Model.
A model that does not contain all the main effects and interactions. In DOE++, a reduced model is the model that does not contain all the effects specified by the user. See also Full Model.
Line 312: Line 313:
   
   


Regression Model
'''Regression Model'''


A model that attempts to explain the relationship between two or more variables.
A model that attempts to explain the relationship between two or more variables.
Line 318: Line 319:
   
   


Repeated Runs
'''Repeated Runs'''


Experiment runs corresponding to the same treatment that are conducted at the same time.
Experiment runs corresponding to the same treatment that are conducted at the same time.
Line 324: Line 325:
   
   


Replicated Runs
'''Replicated Runs'''


Experiment runs corresponding to the same treatment that are conducted in a random order.
Experiment runs corresponding to the same treatment that are conducted in a random order.
Line 330: Line 331:
   
   


Residual
'''Residual'''


An estimate of error which is obtained by calculating the difference between an observation and the corresponding fitted value. See also Error, Fitted Value.
An estimate of error which is obtained by calculating the difference between an observation and the corresponding fitted value. See also Error, Fitted Value.
Line 336: Line 337:
   
   


Residual Sum of Squares
'''Residual Sum of Squares'''


See Error Sum of Squares.
See Error Sum of Squares.
Line 342: Line 343:
   
   


Response
'''Response'''


The quantity that is investigated in an experiment to see which of the factors affect it.
The quantity that is investigated in an experiment to see which of the factors affect it.
Line 348: Line 349:
   
   


Robust System
'''Robust System'''


A system that is insensitive to the effects of noise factors.
A system that is insensitive to the effects of noise factors.
Line 354: Line 355:
   
   


Rotatable Design
'''Rotatable Design'''


A design is rotatable if the variance of the predicted response at any point depends only on the distance of the point from the design center point.
A design is rotatable if the variance of the predicted response at any point depends only on the distance of the point from the design center point.
Line 360: Line 361:
   
   


Screening Designs
'''Screening Designs'''


Experiments that use only a few runs to filter out important main effects and lower order interactions by assuming that higher order interactions are unimportant.
Experiments that use only a few runs to filter out important main effects and lower order interactions by assuming that higher order interactions are unimportant.
Line 366: Line 367:
   
   


Sequential Sum of Squares
'''Sequential Sum of Squares'''


The type of extra sum of squares that is calculated assuming that all terms preceding the given term are included in the model. See also Extra Sum of Squares, Partial Sum of Squares.
The type of extra sum of squares that is calculated assuming that all terms preceding the given term are included in the model. See also Extra Sum of Squares, Partial Sum of Squares.
Line 372: Line 373:
   
   


Signal to Noise Ratio
'''Signal to Noise Ratio'''


The ratios defined by Taguchi to measure variation in the response caused by the noise factors.
The ratios defined by Taguchi to measure variation in the response caused by the noise factors.
Line 378: Line 379:
   
   


Standard Order
'''Standard Order'''


The order of the treatments such that factors are introduced one by one with each new factor being combined with the preceding terms.
The order of the treatments such that factors are introduced one by one with each new factor being combined with the preceding terms.
Line 384: Line 385:
   
   


Sum of Squares
'''Sum of Squares'''


The quantity that is used to measure either a part or all of the variation in a data set.
The quantity that is used to measure either a part or all of the variation in a data set.
Line 390: Line 391:
   
   


Total Sum of Squares
'''Total Sum of Squares'''


The sum of squares that represent all of the variation in a data set.
The sum of squares that represent all of the variation in a data set.
Line 396: Line 397:
   
   


Transformation
'''Transformation'''


The mathematical function that makes the data follow a given characteristic. In the analysis of experiments transformation is used on the response data to make it follow the normal distribution.
The mathematical function that makes the data follow a given characteristic. In the analysis of experiments transformation is used on the response data to make it follow the normal distribution.
Line 402: Line 403:
   
   


Treatment
'''Treatment'''


The levels of a factor in a single factor experiment are also referred to as treatments. In experiments with many factors a combination of the levels of the factors is referred to as a treatment.
The levels of a factor in a single factor experiment are also referred to as treatments. In experiments with many factors a combination of the levels of the factors is referred to as a treatment.
Line 408: Line 409:
   
   


Word
'''Word'''


The effect used in the defining relation. For example, for the defining relation , the word is .
The effect used in the defining relation. For example, for the defining relation , the word is .

Revision as of 00:37, 2 July 2011

Glossary

Alias

Two or more effects are said to be aliased in an experiment if these effects cannot be distinguished from each other. This happens when the columns of the design matrix corresponding to these effects are identical. As a result, the aliased effects are estimated by the same linear combination of observations instead of each effect being estimated by a unique combination.


ANOVA

ANOVA is the acronym for Analysis of Variance. It refers to the procedure of splitting the variability of a data set to conduct various significance tests.


ANOVA Model

The regression model where all factors are treated as qualitative factors. ANOVA models are used in the analysis of experiments to identify significant factor(s) by investigating each level of the factor(s) individually.


Balanced Design

An experiment in which equal number of observations are taken for each treatment.


Blocking

Separation of experiment runs based on the levels of a nuisance factor. Blocking is used to deal with known nuisance factors. You should block what you can and randomize what you cannot. See also Nuisance Factors, Randomization.


Center Point

The experiment run that corresponds to the mid-level of all the factor ranges.


Coded Values

The factor values that are such that the upper limit of the investigated range of the factor becomes and the lower limit becomes . Using coded values makes the experiments with all factors at two levels orthogonal.


Confidence Interval

A closed interval where a certain percentage of the population is likely to lie. For example, a 90% confidence interval with a lower limit of and an upper limit of implies that 90% of the population lies between the values of and .


Confounding

Confounding occurs in a design when certain effects cannot be distinguished from the block effect. This happens when full factorial designs are run using incomplete blocks. In such designs the same linear combination of observations estimates the block effect and the confounded effects. See also Incomplete Blocks.


Contrast

Any linear combination of two or more factor level means such that the coefficients in the combination add up to zero. The difference between the means at any two levels of a factor is an example of a contrast.


Control Factors

The factors affecting the response that are easily manipulated and set by the operator. See also Noise Factors.


Cross Array Design

The experiment design in which every treatment of the inner array is replicated for each run of the outer array. See also Inner Array, Outer Array.


Curvature Test

The test that investigates if the relation between the response and the factors is linear by using center points. See also Center Point.


Defining Relation

For two level fractional factorial experiments, the equation that is used to obtain the fraction from the full factorial experiment. The equation shows which of the columns of the design matrix in the fraction are identical to the first column. For example, the defining relation can be used to obtain a half-fraction of the two level full factorial experiment with three factors , and . The effect(s) used in the equation is called the generator or word.


Degrees of Freedom

The number of independent observations made in excess of the unknowns.


Design Matrix

The matrix whose columns correspond to the levels of the variables (and their interactions) at which observations are recorded.


Design Resolution

The number of factors in the smallest word in a defining relation. Design resolution indicates the degree of aliasing in a fractional factorial design. See also Defining Relation, Word.


Error

The natural variations that occur in a process, even when all the factors are maintained at the same level. See also Residual.


Error Sum of Squares

The variation in the data not captured by the model. The error sum of squares is also called the residual sum of squares. See also Model Sum of Squares, Total Sum of Squares.


Extra Sum of Squares

The increase in the model sum of squares when a term is added to the model.


Factorial Experiment

The experiment in which all combinations of the factor levels are run.


Fractional Factorial Experiment

The experiment where only a fraction of the combinations of the factor levels are run.


Factor

The entity whose effect on the response is investigated in the experiment.


Fitted Value

The estimate of an observation obtained using the model that has been fit to all the observations.


Fixed Effects Model

The ANOVA model used in the experiments where only a limited number of the factor levels are of interest to the experimenter. See also Random Effects Model.


Full Model

The model that includes all the main effects and their interactions. In DOE++, a full model is the model that contains all the effects that are specified by the user. See also Reduced Model.


Generator

See Word.


Hierarchical Model

In DOE++, a model is said to be hierarchical, if, corresponding to every interaction, the main effects of the related factors are included in the model.


Incomplete Blocks

Blocks that do not contain all the treatments of a factorial experiment.


Inner Array

The experiment design used to investigate the control factors under Taguchi's philosophy to design a robust system. See also Robust System, Outer Array, Cross Array.


Interactions

Interaction between factors means that the effect produced by a change in a factor on the response depends on the level of the other factor(s).


Lack-of-Fit Sum of Squares

The portion of the error sum of squares that represents variation in the data not captured because of using a reduced model. See also Reduced Model, Pure Error Sum of Squares.


Least Squares Means

The predicted mean response value for a given factor level while the remaining factors in the model are set to the coded value of zero.


Level

The setting of a factor used in the experiment.


Main Effect

The change in the response due to a change in the level of a factor.


Mean Square

The sum of squares divided by the respective degrees of freedom.


Model Sum of Squares

The portion of the total variability in the data that is explained by the model. See also Error Sum of Squares, Total Sum of Squares.


Multicollinearity

A model with strong dependencies between the independent variables is said to have multicollinearity.


New Observations

Observations that are not part of the data set used to fit the model.


Noise Factors

Those nuisance factors that vary uncontrollably or naturally and can only be controlled for experimental purposes. For example, ambient temperature, atmospheric pressure and humidity are examples of noise factors.


Nuisance Factors

Factors that have an effect on the response but are not of primary interest to the investigator.


Orthogonal Array

An array in which all the columns are orthogonal to each other. Two columns are said to be orthogonal if the sum of the terms resulting from the product of the columns is zero.


Orthogonal Design

An experiment design is orthogonal if the corresponding design matrix is such that the sum of the terms resulting from the product of any two columns is zero. In orthogonal designs the analysis of an effect does not depend on what other effects are included in the model.


Outer Array

The experiment design used to investigate noise factors under Taguchi's philosophy to design a robust system. See also Robust System, Inner Array, Cross Array.


Partial Sum of Squares

The type of extra sum of squares that is calculated assuming that all terms other than the given term are included in the model. The partial sum of squares is also referred to as the adjusted sum of squares. See also Extra Sum of Squares, Sequential Sum of Squares.


Prediction Interval

The confidence interval on new observations.


Pure Error Sum of Squares

The portion of the error sum of squares that represents variation due to replicates. See also Lack-of-Fit Sum of Squares.


Qualitative Factor

The factor where the levels represent different categories and no numerical ordering is implied. These factors are also called categorical factors.


Random Effects Model

The ANOVA model used in the experiments where the factor levels to be investigated are randomly selected from a large or infinite population. See also Fixed Effects Model.


Randomization

Conducting experiment runs in a random order to cancel out the effect of unknown nuisance factors. See also Blocking.


Randomized Complete Block Design

An experiment design where each block contains one replicate of the experiment and runs within the block are subjected to randomization.


Reduced Model

A model that does not contain all the main effects and interactions. In DOE++, a reduced model is the model that does not contain all the effects specified by the user. See also Full Model.


Regression Model

A model that attempts to explain the relationship between two or more variables.


Repeated Runs

Experiment runs corresponding to the same treatment that are conducted at the same time.


Replicated Runs

Experiment runs corresponding to the same treatment that are conducted in a random order.


Residual

An estimate of error which is obtained by calculating the difference between an observation and the corresponding fitted value. See also Error, Fitted Value.


Residual Sum of Squares

See Error Sum of Squares.


Response

The quantity that is investigated in an experiment to see which of the factors affect it.


Robust System

A system that is insensitive to the effects of noise factors.


Rotatable Design

A design is rotatable if the variance of the predicted response at any point depends only on the distance of the point from the design center point.


Screening Designs

Experiments that use only a few runs to filter out important main effects and lower order interactions by assuming that higher order interactions are unimportant.


Sequential Sum of Squares

The type of extra sum of squares that is calculated assuming that all terms preceding the given term are included in the model. See also Extra Sum of Squares, Partial Sum of Squares.


Signal to Noise Ratio

The ratios defined by Taguchi to measure variation in the response caused by the noise factors.


Standard Order

The order of the treatments such that factors are introduced one by one with each new factor being combined with the preceding terms.


Sum of Squares

The quantity that is used to measure either a part or all of the variation in a data set.


Total Sum of Squares

The sum of squares that represent all of the variation in a data set.


Transformation

The mathematical function that makes the data follow a given characteristic. In the analysis of experiments transformation is used on the response data to make it follow the normal distribution.


Treatment

The levels of a factor in a single factor experiment are also referred to as treatments. In experiments with many factors a combination of the levels of the factors is referred to as a treatment.


Word

The effect used in the defining relation. For example, for the defining relation , the word is .