In statistics, a Yates analysis is an approach to analyzing data obtained from a designed experiment, where a factorial design has been used. Full- and fractional-factorial designs are common in designed experiments for engineering and scientific applications. In these designs, each factor is assigned two levels, typically called the low and high levels, and referred to as "-" and "+". For computational purposes, the factors are scaled so that the low level is assigned a value of -1 and the high level is assigned a value of +1.
A full factorial design contains all possible combinations of low/high levels for all the factors. A fractional factorial design contains a carefully chosen subset of these combinations. The criterion for choosing the subsets is discussed in detail in the fractional factorial designs article.
Formalized by Frank Yates, a Yates analysis exploits the special structure of these designs to generate least squares estimates for factor effects for all factors and all relevant interactions. The Yates analysis can be used to answer the following questions:
What is the ranked list of factors?
What is the goodness-of-fit (as measured by the residual standard deviation) for the various models?
The mathematical details of the Yates analysis are given in chapter 10 of Box, Hunter, and Hunter (1978).
Before performing a Yates analysis, the data should be arranged in "Yates' order". That is, given k factors, the kth column consists of 2(k - 1) minus signs (i.e., the low level of the factor) followed by 2(k - 1) plus signs (i.e., the high level of the factor). For example, for a full factorial design with three factors, the design matrix is
To better understand and utilize the sign table above, one method of detailing the factors and treatment combinations is called modern notation. The notation is a shorthand that arises from taking the subscripts of each treatment combination, making them exponents, and then evaluating the resulting expression and using that as the new name of the treatment combination. Note that while the names look very much like algebraic expressions, they are simply names and no new values are assigned. Taking the 3-factor, 2-level model from above, in Yates' order the response variables are:
which in modern notation becomes:
in which it is evident that the exponents in the modern notation names are simply the subscripts of the former (note that anything raised to the zeroth power is 1 and anything raised to the first power is itself). Each response variable then gets assigned row-wise to the table above. Thus, the first row is for , the second row is for , and so on. The signs in each column then represent the signs each response variable should take in the calculation of the effect estimates for that factor.
Determining the Yates' order for fractional factorial designs requires knowledge of the confounding structure of the fractional factorial design.
Output
A Yates analysis generates the following output.
A factor identifier (from Yates' order). The specific identifier will vary depending on the program used to generate the Yates analysis. Dataplot, for example, uses the following for a 3-factor model.
1 = factor 1
2 = factor 2
3 = factor 3
12 = interaction of factor 1 and factor 2
13 = interaction of factor 1 and factor 3
23 = interaction of factor 2 and factor 3
123 = interaction of factors 1, 2, and 3
A ranked list of important factors. That is, least squares estimated factor effects ordered from largest in magnitude (most significant) to smallest in magnitude (least significant).
To determine the magnitudes, the response variables are first arranged in Yates' order, as described in the aforementioned section above. Then, terms are added and subtracted pairwise to determine the next column. More specifically, given the values of the response variables (as they should have been obtained from the experiment directly) in Yates' order, the first two terms are added and that sum is now the first term in the new column. The next two terms are then added and that is the second term in the new column. Since the terms are added pairwise, half of the new column is now filled and should be comprised entirely of the pairwise sums of the data. The second half of the column is found in an analogous manner, only the pairwise differences are taken, where the first term is subtracted from the second, the third from the fourth, and so on. Thus completes the column. Should more columns be needed, the same process is repeated, only using the new column. In other words, the nth column is generated from the (n-1)th column (Berger et al. calls this process "Yatesing the data"). In a design, k columns will be required, and the last column is the column used to calculate the effect estimates.
A t-value for the individual factor effect estimates. The t-value is computed as
where e is the estimated factor effect and se is the standard deviation of the estimated factor effect.
The residual standard deviation that results from the model with the single term only. That is, the residual standard deviation from the model
where Xi is the estimate of the ith factor or interaction effect.
The cumulative residual standard deviation that results from the model using the current term plus all terms preceding that term. That is,
This consists of a monotonically decreasing set of residual standard deviations (indicating a better fit as the number of terms in the model increases). The first cumulative residual standard deviation is for the model
where the constant is the overall mean of the response variable. The last cumulative residual standard deviation is for the model
This last model will have a residual standard deviation of zero.
Example
(Adapted from Berger et al., chapter 9) Say there is a study done where one is selling some product by mail and is trying to determine the effect of three factors (postage, product price, envelope size) on people's response rate (that is, will they buy the product). Each factor has two levels: for postage (labeled A), they are third-class (low) and first-class (high), for product price (labeled B) the low level is $9.95 and the high level is $12.95, and for envelope size (labeled C) the low level is #10 and the high level is 9x12. From the experiment, the following data are acquired. Note that the response rate is given as a proportion of the people who answered the survey (favorably and unfavorably) for each treatment combination.
Response rate for treatment combination
Treatmemt combination
Response rate
1
0.062
a
0.074
b
0.010
ab
0.020
c
0.057
ac
0.082
bc
0.024
abc
0.027
Singling out factor A (postage) for calculation for now, the overall estimate for A must take into account the interaction effects of B and C on it as well. The four terms for calculation are:
a –1, estimate of A with B and C low
ab – b, estimate of A with B high and C low
ac – c, estimate of A with B low and C high
abc – bc, estimate of A with B and C high
The total estimate is the sum of these four terms divided by four. In other words, the estimate of A is
where the sum has been rearranged to have all the positive terms grouped together and the negatives together for ease of viewing. In Yates' order, the sum is written as
The estimates for B and C can be determined in a similar fashion. Calculating the interaction effects is also similar, but the responses are averaged over all other effects not considered.
Table of signs to calculate the effect estimates for a 3-level, 2-factor factorial design. Adapted from Berger et al., ch. 9
The full table of signs for a three-factor, two-level design is given to the right. Both the factors (columns) and the treatment combinations (rows) are written in Yates' order. The value of arranging the sum in Yates' order is now apparent, as only the signs need to be altered according to the table to produce the effect estimates for every treatment combination. Observe that the columns for A, B, and C are the same as those in the design matrix in the above Yates' order section. Observe also that the columns of the interaction effects can be produced by taking the dot product of the columns of the individual factors (i.e., multiplying the columns element-wise to produce another column of the same length). Note that all sums must divided by 4 to yield the actual effect estimate, as shown earlier. Using this table, the effect estimates are calculated as:
Effect estimate for treatment combination
Treatmemt combination
Effect estimate
A
0.0125
B
–0.0485
AB
–0.0060
C
0.0060
AC
0.0015
BC
0.0045
ABC
–0.0050
A positive value means that an increase in the factor creates an increase in the response rate, while a negative value means that that same factor increase actually produces a decrease in the response rate. Note however that these effects have not yet been determined to be statistically significant, only that there are such effects on the response rate for each factor. Statistical significance must be ascertained via other methods such as analysis of variance (ANOVA).
Further reading can be found in Berger et al., chapter 9.
Parameter estimates as terms are added
In most cases of least squares fitting, the model coefficients for previously added terms change depending on what was successively added. For example, the X1 coefficient might change depending on whether or not an X2 term was included in the model. This is not the case when the design is orthogonal, as is a 23 full factorial design. For orthogonal designs, the estimates for the previously included terms do not change as additional terms are added. This means the ranked list of effect estimates simultaneously serves as the least squares coefficient estimates for progressively more complicated models.
Model selection and validation
From the above Yates output, one can define the potential models from the Yates analysis. An important component of a Yates analysis is selecting the best model from the available potential models. The above step lists all the potential models. From this list, we want to select the most appropriate model. This requires balancing the following two goals.
We want the model to include all important factors.
We want the model to be parsimonious. That is, the model should be as simple as possible.
In short, we want our model to include all the important factors and interactions and to omit the unimportant factors and interactions. Note that the residual standard deviation alone is insufficient for determining the most appropriate model as it will always be decreased by adding additional factors. Instead, seven criteria are utilized to define important factors. These seven criteria are not all equally important, nor will they yield identical subsets, in which case a consensus subset or a weighted consensus subset must be extracted. In practice, some of these criteria may not apply in all situations, and some analysts may have additional criteria. These criteria are given as useful guidelines. Mosts analysts will focus on those criteria that they find most useful.
The first four criteria focus on effect sizes with three numeric criteria and one graphical criterion. The fifth criterion focuses on averages. The last two criteria focus on the residual standard deviation of the model. Once a tentative model has been selected, the error term should follow the assumptions for a univariate measurement process. That is, the model should be validated by analyzing the residuals.
Graphical presentation
Some analysts may prefer a more graphical presentation of the Yates results. In particular, the following plots may be useful:
Analysis of variance (ANOVA) is a collection of statistical models and their associated estimation procedures used to analyze the differences among means. ANOVA was developed by the statistician Ronald Fisher. ANOVA is based on the law of total variance, where the observed variance in a particular variable is partitioned into components attributable to different sources of variation. In its simplest form, ANOVA provides a statistical test of whether two or more population means are equal, and therefore generalizes the t-test beyond two means. In other words, the ANOVA is used to test the difference between two or more means.
Analysis of covariance (ANCOVA) is a general linear model that blends ANOVA and regression. ANCOVA evaluates whether the means of a dependent variable (DV) are equal across levels of one or more categorical independent variables (IV) and across one or more continuous variables. For example, the categorical variable(s) might describe treatment and the continuous variable(s) might be covariates or nuisance variables; or vice versa. Mathematically, ANCOVA decomposes the variance in the DV into variance explained by the CV(s), variance explained by the categorical IV, and residual variance. Intuitively, ANCOVA can be thought of as 'adjusting' the DV by the group means of the CV(s).
In statistics, an interaction may arise when considering the relationship among three or more variables, and describes a situation in which the effect of one causal variable on an outcome depends on the state of a second causal variable. Although commonly thought of in terms of causal relationships, the concept of an interaction can also describe non-causal associations. Interactions are often considered in the context of regression analyses or factorial experiments.
The normal probability plot is a graphical technique to identify substantive departures from normality. This includes identifying outliers, skewness, kurtosis, a need for transformations, and mixtures. Normal probability plots are made of raw data, residuals from model fits, and estimated parameters.
In statistics, a full factorial experiment is an experiment whose design consists of two or more factors, each with discrete possible values or "levels", and whose experimental units take on all possible combinations of these levels across all such factors. A full factorial design may also be called a fully crossed design. Such an experiment allows the investigator to study the effect of each factor on the response variable, as well as the effects of interactions between factors on the response variable.
This glossary of statistics and probability is a list of definitions of terms and concepts used in the mathematical sciences of statistics and probability, their sub-disciplines, and related fields. For additional related terms, see Glossary of mathematics and Glossary of experimental design.
Multilevel models are statistical models of parameters that vary at more than one level. An example could be a model of student performance that contains measures for individual students as well as measures for classrooms within which the students are grouped. These models can be seen as generalizations of linear models, although they can also extend to non-linear models. These models became much more popular after sufficient computing power and software became available.
In statistics, a central composite design is an experimental design, useful in response surface methodology, for building a second order (quadratic) model for the response variable without needing to use a complete three-level factorial experiment.
In statistics, fractional factorial designs are experimental designs consisting of a carefully chosen subset (fraction) of the experimental runs of a full factorial design. The subset is chosen so as to exploit the sparsity-of-effects principle to expose information about the most important features of the problem studied, while using a fraction of the effort of a full factorial design in terms of experimental runs and resources. In other words, it makes use of the fact that many experiments in full factorial design are often redundant, giving little or no new information about the system.
In the design of experiments and analysis of variance, a main effect is the effect of an independent variable on a dependent variable averaged across the levels of any other independent variables. The term is frequently used in the context of factorial designs and regression models to distinguish main effects from interaction effects.
In computational biology and bioinformatics, analysis of variance – simultaneous component analysis is a method that partitions variation and enables interpretation of these partitions by SCA, a method that is similar to principal components analysis (PCA). Analysis of variance (ANOVA) is a collection of statistical models and their associated estimation procedures used to analyze differences. Statistical coupling analysis (SCA) is a technique used in bioinformatics to measure covariation between pairs of amino acids in a protein multiple sequence alignment (MSA).
In statistics, one-way analysis of variance is a technique to compare whether two or more samples' means are significantly different. This analysis of variance technique requires a numeric response variable "Y" and a single explanatory variable "X", hence "one-way".
An alpha beta filter is a simplified form of observer for estimation, data smoothing and control applications. It is closely related to Kalman filters and to linear state observers used in control theory. Its principal advantage is that it does not require a detailed system model.
In statistics, restricted randomization occurs in the design of experiments and in particular in the context of randomized experiments and randomized controlled trials. Restricted randomization allows intuitively poor allocations of treatments to experimental units to be avoided, while retaining the theoretical benefits of randomization. For example, in a clinical trial of a new proposed treatment of obesity compared to a control, an experimenter would want to avoid outcomes of the randomization in which the new treatment was allocated only to the heaviest patients.
A glossary of terms used in experimental research.
In statistics and regression analysis, moderation occurs when the relationship between two variables depends on a third variable. The third variable is referred to as the moderator variable or simply the moderator. The effect of a moderating variable is characterized statistically as an interaction; that is, a categorical or continuous variable that is associated with the direction and/or magnitude of the relation between dependent and independent variables. Specifically within a correlational analysis framework, a moderator is a third variable that affects the zero-order correlation between two other variables, or the value of the slope of the dependent variable on the independent variable. In analysis of variance (ANOVA) terms, a basic moderator effect can be represented as an interaction between a focal independent variable and a factor that specifies the appropriate conditions for its operation.
In statistics, the two-way analysis of variance (ANOVA) is an extension of the one-way ANOVA that examines the influence of two different categorical independent variables on one continuous dependent variable. The two-way ANOVA not only aims at assessing the main effect of each independent variable but also if there is any interaction between them.
A robust parameter design, introduced by Genichi Taguchi, is an experimental design used to exploit the interaction between control and uncontrollable noise variables by robustification—finding the settings of the control factors that minimize response variation from uncontrollable factors. Control variables are variables of which the experimenter has full control. Noise variables lie on the other side of the spectrum. While these variables may be easily controlled in an experimental setting, outside of the experimental world they are very hard, if not impossible, to control. Robust parameter designs use a naming convention similar to that of FFDs. A 2(m1+m2)-(p1-p2) is a 2-level design where m1 is the number of control factors, m2 is the number of noise factors, p1 is the level of fractionation for control factors, and p2 is the level of fractionation for noise factors.
In statistics, linear regression is a statistical model which estimates the linear relationship between a scalar response and one or more explanatory variables. The case of one explanatory variable is called simple linear regression; for more than one, the process is called multiple linear regression. This term is distinct from multivariate linear regression, where multiple correlated dependent variables are predicted, rather than a single scalar variable. If the explanatory variables are measured with error then errors-in-variables models are required, also known as measurement error models.
This page is based on this Wikipedia article Text is available under the CC BY-SA 4.0 license; additional terms may apply. Images, videos and audio are available under their respective licenses.
