Each treatment is defined by the choice of one of two levels for each of the four factors. In the R code above, we have used the function FrF2 (from the package of the same name) to generate all \(t = 2^4 = 16\) combinations of the two levels of these four factors. We come back to this function later in the chapter.
This factorial treatment structure lends itself to certain treatment contrasts being of natural interest.
Throughout this chapter, we will assume there are no blocks or other restrictions on randomisation, and so we will assume a completely randomised design can be used. We start by assuming the same unit-treatment model as Chapter 2 :
\[\begin{equation} y_{ij} = \mu + \tau_i + \varepsilon_{ij}\,, \quad i = 1, \ldots, t; j = 1, \ldots, n_i\,, \tag{4.1} \end{equation}\]
where \(y_{ij}\) is the response from the \(j\) th application of treatment \(i\) , \(\mu\) is a constant parameter, \(\tau_i\) is the effect of the \(i\) th treatment, and \(\varepsilon_{ij}\) is the random individual effect from each experimental unit with \(\varepsilon_{ij} \sim N(0, \sigma^2)\) independent of other errors.
Now, the number of treatments \(t = 2^f\) , where \(f\) equals the number of factors in the experiment.
For Example 4.1 , we have \(t = 2^4 = 16\) and \(n_i = 1\) for all \(i=1,\ldots, 16\) ; that is, each of the 16 treatments are replicated once. In general, we shall assume common treatment replication \(n_i = r \ge 1\) .
If we fit model (4.1) and compute the ANOVA table, we notice a particular issue with this design.
All available degrees of freedom are being used to estimate parameters in the mean ( \(\mu\) and the treatment effects \(\tau_i\) ). There are no degrees of freedom left to estimate \(\sigma^2\) . This is due to a lack of treatment replication. Without replication in the design, model (4.1) is saturated , with as many treatments as there are observations and an unbiased estimate of \(\sigma^2\) cannot be obtained. We will return to this issue later.
Studying Table 4.1 , there are some comparisons between treatments which are obviously of interest. For example, comparing the average effect from the first 8 treatments with the average effect of the second 8, using
\[ \boldsymbol{c}^{\mathrm{T}}\boldsymbol{\tau} = \sum_{i=1}^tc_i\tau_i\,, \] with
\[ \boldsymbol{c}^{\mathrm{T}} = (-\boldsymbol{1}_{2^{f-1}}^{\mathrm{T}}, \boldsymbol{1}_{2^{f-1}}^{\mathrm{T}}) / 2^{f-1} = (-\boldsymbol{1}_8^{\mathrm{T}}, \boldsymbol{1}_8^{\mathrm{T}}) / 8\,. \]
This contrast compares the average treatment effect from the 8 treatments which have reagent set to its low level (1 equiv.) to the average effect from the 8 treatments which have reagent set to its high level. This is a “fair” comparison, as both of these sets of treatments have each of the combinations of the factors temp , time and solvent occuring equally often (twice here). Hence, the main effect of reagent is averaged over the levels of the other three factors.
As in Chapter 2 , we can estimate this treatment contrast by applying the same contrast coefficients to the treatment means,
\[ \widehat{\boldsymbol{c}^{\mathrm{T}}\boldsymbol{\tau}} = \sum_{i=1}^tc_i\bar{y}_{i.}\,, \] where, for this experiment, each \(\bar{y}_{i.}\) is the mean of a single observation (as there is no treatment replication). We see that inference about this contrast is not possible, as no standard error can be obtained.
Definition 4.1 The main effect of a factor \(A\) is defined as the difference in the average response from the high and low levels of the factor
\[ \mbox{ME}(A) = \bar{y}(A+) - \bar{y}(A-)\,, \] where \(\bar{y}(A+)\) is the average response when factor \(A\) is set to its high level, averaged across all combinations of levels of the other factors (with \(\bar{y}(A+)\) defined similarly for the low level of \(A\) ).
As we have averaged the response across the levels of the other factors, the intepretation of the main effect extends beyond this experiment. That is, we can use it to infer something about the system under study. Assuming model (4.1) is correct, any variation in the main effect can only come from random error in the observations. In fact,
\[\begin{align*} \mbox{var}\{ME(A)\} & = \frac{\sigma^2}{n/2} + \frac{\sigma^2}{n/2} \\ & = \frac{4\sigma^2}{n}\,, \end{align*}\]
and assuming \(r>1\) ,
\[\begin{equation} \hat{\sigma}^2 = \frac{1}{2^f(r-1)} \sum_{i=1}^{2^f}\sum_{j=1}^r(y_{ij} - \bar{y}_{i.})^2\,, \tag{4.2} \end{equation}\]
which is the residual mean square.
For Example 4.1 , we can also calculate main effect estimates for the other three factors by defining appropriate contrasts in the treatments.
Temperature | Time | Solvent | Reagent | |
---|---|---|---|---|
Trt 1 | -0.125 | -0.125 | -0.125 | -0.125 |
Trt 2 | 0.125 | -0.125 | -0.125 | -0.125 |
Trt 3 | -0.125 | 0.125 | -0.125 | -0.125 |
Trt 4 | 0.125 | 0.125 | -0.125 | -0.125 |
Trt 5 | -0.125 | -0.125 | 0.125 | -0.125 |
Trt 6 | 0.125 | -0.125 | 0.125 | -0.125 |
Trt 7 | -0.125 | 0.125 | 0.125 | -0.125 |
Trt 8 | 0.125 | 0.125 | 0.125 | -0.125 |
Trt 9 | -0.125 | -0.125 | -0.125 | 0.125 |
Trt 10 | 0.125 | -0.125 | -0.125 | 0.125 |
Trt 11 | -0.125 | 0.125 | -0.125 | 0.125 |
Trt 12 | 0.125 | 0.125 | -0.125 | 0.125 |
Trt 13 | -0.125 | -0.125 | 0.125 | 0.125 |
Trt 14 | 0.125 | -0.125 | 0.125 | 0.125 |
Trt 15 | -0.125 | 0.125 | 0.125 | 0.125 |
Trt 16 | 0.125 | 0.125 | 0.125 | 0.125 |
Estimates can be obtained by applying these coefficients to the observed treatment means.
Main effects are often displayed graphically, using main effect plots which simply plot the average response for each factor level, joined by a line. The larger the main effect, the larger the slope of the line (or the bigger the difference between the averages). Figure 4.1 presents the four main effect plots for Example 4.1 .
Figure 4.1: Desilylation experiment: main effect plots
Another contrast that could be of interest in Example 4.1 has coefficients
\[ \boldsymbol{c}^{\mathrm{T}} = (\boldsymbol{1}_4^{\mathrm{T}}, -\boldsymbol{1}_8^{\mathrm{T}}, \boldsymbol{1}_4^{\mathrm{T}}) / 8 \,, \] where the divisor \(8 = 2^{f-1} = 2^3\) .
This contrast measures the difference between the average treatment effect from treatments 1-4, 13-16 and treatments 5-12. Checking back against Table 4.1 , we see this is comparing those treatments where solvent and reagent are both set to their low (1-4) or high (13-16) level against those treatments where one of the two factors is set high and the other is set low (5-12).
Focusing on reagent , if the effect of this factor on the response was independent of the level to which solvent has been set, you would expect this contrast to be zero - changing from the high to low level of reagent should affect the response in the same way, regardless of the setting of solvent . This argument can be reversed, focussing on the effect of solvent . Therefore, if this contrast is large, we say the two factors interact .
For Example 4.1 , this interaction contrast seems quite small, although of course without an estimate of the standard error we are still lacking a formal method to judge this.
It is somewhat more informative to consider the above interaction contrast as the average difference in two “sub-contrasts”
\[ \boldsymbol{c}^{\mathrm{T}}\boldsymbol{\tau} = \frac{1}{2}\left\{\frac{1}{4}\left(\tau_{13} + \tau_{14} + \tau_{15} + \tau_{16} - \tau_5 - \tau_6 - \tau_7 - \tau_8\right) - \frac{1}{4}\left(\tau_9 + \tau_{10} + \tau_{11} + \tau_{12} - \tau_1 - \tau_2 - \tau_3 - \tau_4\right) \right\}\,, \] The first component in the above expression is the effect of changing reagent from high to low given solvent is set to it’s high level . The second component the effect of changing reagent from high to low given solvent is set to it’s low level . This leads to our definition of a two-factor interaction.
Definition 4.2 The two-factor interaction between factors \(A\) and \(B\) is defined as the average difference in main effect of factor \(A\) when computed at the high and low levels of factor \(B\) .
\[\begin{align*} \mbox{Int}(A, B) & = \frac{1}{2}\left\{\mbox{ME}(A\mid B+) - \mbox{ME}(A \mid B-)\right\} \\ & = \frac{1}{2}\left\{\mbox{ME}(B \mid A+) - \mbox{ME}(B \mid A-)\right\} \\ & = \frac{1}{2}\left\{\bar{y}(A+, B+) - \bar{y}(A-, B+) - \bar{y}(A+, B-) + \bar{y}(A-, B-)\right\}\,, \end{align*}\]
where \(\bar{y}(A+, B-)\) is the average response when factor \(A\) is set to its high level and factor \(B\) is set to its low level, averaged across all combinations of levels of the other factors, and other averages are defined similarly. The conditional main effects of factor \(A\) when factor \(B\) is set to its high level is defined as
\[ \mbox{ME}(A\mid B+) = \bar{y}(A+, B+) - \bar{y}(A-, B+)\,, \]
with similar definitions for other conditional main effects.
As the sum of the squared contrast coefficients is the same for two-factor interactions as for main effects, the variance of the contrast estimator is also the same.
\[ \mbox{var}\left\{\mbox{Int}(A, B)\right\} = \frac{4\sigma^2}{n}\,. \] For Example 4.1 we can calculate two-factor interactions for all \({4 \choose 2} = 6\) pairs of factors. The simplest way to calculate the contrast coefficients is as the elementwise, or Hadamard, product 25 of the unscaled main effect contrasts (before dividing by \(2^{f-1}\) ).
tem_x_tim | tem_x_sol | tem_x_rea | tim_x_sol | tim_x_rea | sol_x_rea | |
---|---|---|---|---|---|---|
Trt 1 | 0.125 | 0.125 | 0.125 | 0.125 | 0.125 | 0.125 |
Trt 2 | -0.125 | -0.125 | -0.125 | 0.125 | 0.125 | 0.125 |
Trt 3 | -0.125 | 0.125 | 0.125 | -0.125 | -0.125 | 0.125 |
Trt 4 | 0.125 | -0.125 | -0.125 | -0.125 | -0.125 | 0.125 |
Trt 5 | 0.125 | -0.125 | 0.125 | -0.125 | 0.125 | -0.125 |
Trt 6 | -0.125 | 0.125 | -0.125 | -0.125 | 0.125 | -0.125 |
Trt 7 | -0.125 | -0.125 | 0.125 | 0.125 | -0.125 | -0.125 |
Trt 8 | 0.125 | 0.125 | -0.125 | 0.125 | -0.125 | -0.125 |
Trt 9 | 0.125 | 0.125 | -0.125 | 0.125 | -0.125 | -0.125 |
Trt 10 | -0.125 | -0.125 | 0.125 | 0.125 | -0.125 | -0.125 |
Trt 11 | -0.125 | 0.125 | -0.125 | -0.125 | 0.125 | -0.125 |
Trt 12 | 0.125 | -0.125 | 0.125 | -0.125 | 0.125 | -0.125 |
Trt 13 | 0.125 | -0.125 | -0.125 | -0.125 | -0.125 | 0.125 |
Trt 14 | -0.125 | 0.125 | 0.125 | -0.125 | -0.125 | 0.125 |
Trt 15 | -0.125 | -0.125 | -0.125 | 0.125 | 0.125 | 0.125 |
Trt 16 | 0.125 | 0.125 | 0.125 | 0.125 | 0.125 | 0.125 |
Estimates of the interaction contrasts can again by found by considering the equivalent contrasts in the observed treatment means.
As with main effects, interactions are often displayed graphically using interaction plots, plotting average responses for each pairwise combination of factors, joined by lines.
Figure 4.2: Desilylation experiment: two-factor interaction plots
Parallel lines in an interaction plot indicate no (or very small) interaction ( time and solvent , time and reagent , solvent and reagent ). The three interactions with temp all demonstrate much more robust behaviour at the high level; changing time , solvent or reagent makes little difference to the response at the high level of temp , and much less difference than at the low level of temp .
If a system displays important interactions, the main effects of factors involved in those interactions should no longer be interpreted. For example, it makes little sense to discuss the main effect of temp when is changes so much with the level of reagent (from strongly positive when reagent is low to quite small when reagent is high).
Higher order interactions can be defined similarly, as average differences in lower-order effects. For example, a three-factor interaction measures how a two-factor interaction changes with the levels of a third factor.
\[\begin{align*} \mbox{Int}(A, B, C) & = \frac{1}{2}\left\{\mbox{Int}(A, B \mid C+) - \mbox{Int}(A, B \mid C-)\right\} \\ & = \frac{1}{2}\left\{\mbox{Int}(A, C \mid B+) - \mbox{Int}(A, C \mid B-)\right\} \\ & = \frac{1}{2}\left\{\mbox{Int}(B, C \mid A+) - \mbox{Int}(B, C \mid A-)\right\}\,, \\ \end{align*}\]
\[ \mbox{Int}(A, B \mid C+) = \frac{1}{2}\left\{\bar{y}(A+, B+, C+) - \bar{y}(A-, B+, C+) - \bar{y}(A+, B-, C+) + \bar{y}(A-, B-, C+)\right\} \] is the interaction between factors \(A\) and \(B\) using only those treatments where factor \(C\) is set to it’s high level. Higher order interaction contrasts can again be constructed by (multiple) hadamard products of (unscaled) main effect contrasts.
Definition 4.3 A factorial effect is a main effect or interaction contrast defined on a factorial experiment. For a \(2^f\) factorial experiment with \(f\) factors, there are \(2^f-1\) factorial effects, ranging from main effects to the interaction between all \(f\) factors. The contrast coefficients in a factorial contrast all take the form \(c_i = \pm 1 / 2^{f-1}\) .
For Example 4.1 , we can now calculate all the factorial effects.
Empirical study of many experiments ( Box and Meyer, 1986 ; Li et al. , 2006 ) have demonstrated that the following three principles often hold when analysing factorial experiments.
Definition 4.4 Effect hierarchy : lower-order factorial effects are more likely to be important than higher-order effects; factorial effects of the same order are equally likely to be important.
For example, we would anticipate more large main effects from the analysis of a factorial experiment than two-factor interactions.
Definition 4.5 Effect sparsity : the number of large factorial effects is likely to be small, relative to the total number under study.
This is sometimes called the pareto principle .
Definition 4.6 Effect heredity : interactions are more likely to be important if at least one parent factor also has a large main effect.
These three principles will provide us with some useful guidelines when analysing, and eventually constructing, factorial experiments.
The lack of an estmate for \(\sigma^2\) means alternatives to formal inference methods (e.g. hypothesis tests) must be found to assess the size of factorial effects. We will discuss a method that essentially treats the identification of large factorial effects as an outlier identification problem.
Let \(\hat{\theta}_j\) be the \(j\) th estimated factorial effect, with \(\hat{\theta}_j = \sum_{i=1}^tc_{ij}\bar{y}_{i.}\) for \(\boldsymbol{c}_j^{\mathrm{T}} = (c_{1j}, \ldots, c_{tj})\) a vector of factorial contrast coefficients (defining a main effect or interaction). Then the estimator follows a normal distribution
\[ \hat{\theta}_j \sim N\left(\theta_j, \frac{4\sigma^2}{n}\right)\,,\qquad j = 1, \ldots, 2^f-1\,, \] for \(\theta_j\) the true, unknown, value of the factorial effect, \(j = 1,\ldots, 2^f\) . Further more, for \(j, l = 1, \ldots 2^f-1; \, j\ne l\) ,
\[\begin{align*} \mbox{cov}(\hat{\theta}_j, \hat{\theta}_l) & = \mbox{cov}\left(\sum_{i=1}^tc_{ij}\bar{y}_{i.}, \sum_{i=1}^tc_{il}\bar{y}_{i.}\right) \\ & = \sum_{i=1}^tc_{ij}c_{il}\mbox{var}(\bar{y}_{i.}) \\ & = \frac{\sigma^2}{r} \sum_{i=1}^tc_{ij}c_{il} \\ & = 0\,, \\ \end{align*}\]
as \(\sum_{i=1}^tc_{ij}c_{il} = 0\) for \(j\ne l\) . That is, the factorial contrasts are independent as the contrast coefficient vectors are orthogonal.
Hence, under the null hypothesis \(H_0: \theta_1 = \cdots = \theta_{2^f-1} = 0\) (all factorial effects are zero), the \(\hat{\theta}_j\) form a sample from independent normally distributed random variables from the distribution
\[\begin{equation} \hat{\theta}_j \sim N\left(0, \frac{4\sigma^2}{n}\right)\,,\qquad j = 1, \ldots, 2^f-1\,. \tag{4.3} \end{equation}\]
To assess evidence against \(H_0\) , we can plot the ordered estimates of the factorial effects against the ordered quantiles of a standard normal distribution. Under \(H_0\) , the points in this plot should lie on a straightline (the slope of the line will depend on the unknown \(\sigma^2\) ). We anticipate that the majority of the effects will be small ( effect sparsity ), and hence any large effects that lie away from the line are unlikely to come from distribution (4.3) and may be significantly different from zero. We have essentially turned the problem into an outlier identification problem.
For Example 4.1 , we can easily produce this plot in R . Table 4.4 gives the ordered factorial effects, which are then plotted against standard normal quantiles in Figure 4.3 .
contrast | estimate |
---|---|
temp | 8.1200 |
reagent | 3.0875 |
time | 2.5675 |
temp.solvent | 2.3575 |
solvent.reagent | 0.4900 |
time.solvent | 0.4400 |
temp.time.solvent | 0.2450 |
temp.time.reagent | 0.1950 |
temp.time.solvent.reagent | 0.1925 |
temp.solvent.reagent | -0.0300 |
time.solvent.reagent | -0.2375 |
time.reagent | -0.6450 |
solvent | -2.2175 |
temp.time | -2.3575 |
temp.reagent | -2.7725 |
Figure 4.3: Desilylation experiment: normal effects plot
In fact, it is more usual to use a half-normal plot to assess the size of factorial effects, where we plot the sorted absolute values of the estimated effects against the quantiles of a half-normal distribution 26
Figure 4.4: Desilylation experiment: half-normal effects plot
The advantage of a half-normal plot such as Figure 4.4 is that we only need to look at effects appearing in the top right corner (significant effects will always appear “above” a hypothetical straight line) and we do not need to worry about comparing large positive and negative values. For these reason, they are usually preferred over normal plots.
For the desilylation experiment, we can see the effects fall into three groups: one effect standing well away from the line, and almost certainly significant ( temp , from Table 4.4 ), then a group of six effects ( reagent , time , temp.solvent , solvent , temp.time , temp.reagent ) which may be significant, and then a group of 8 much smaller effects.
The assessment of normal or half-normal effect plots can be quite subjective. Lenth ( 1989 ) introduced a simple method for conducting more formal hypothesis testing in unreplicated factorial experiments.
Lenth’s method uses a pseudo standard error (PSE):
\[ \mbox{PSE} = 1.5 \times \mbox{median}_{|\hat{\theta}_i| < 2.5s_0}|\hat{\theta}_i|\,, \] where \(s_0 = 1.5\times \mbox{median} |\hat{\theta}_i|\) is a consistent 27 estimator of the standard deviation of the \(\hat{\theta}_i\) under \(H_0: \theta_1 = \cdots=\theta_{2^f-1}=0\) . The PSE trims approximately 1% 28 of the \(\hat{\theta}_i\) to produce a robust estimator of the standard deviation, in the sense that it is not influenced by large \(\hat{\theta}_i\) belonging to important effects.
For Example 4.1 , we can construct the PSE as follows.
The PSE can be used to construct test statistics
\[ t_{\mbox{PSE}, i} = \frac{\hat{\theta}_i}{\mbox{PSE}}\,, \]
which mimic the usual \(t\) -statistics used when an estimate of \(\sigma^2\) is available. These quantities can be compared to reference distribution which was tabulated by Lenth ( 1989 ) and simulated in R using the unrepx package.
effect | Lenth_PSE | t.ratio | p.value | simult.pval | |
---|---|---|---|---|---|
temp | 8.1200 | 0.66 | 12.303 | 0.0001 | 0.0007 |
reagent | 3.0875 | 0.66 | 4.678 | 0.0039 | 0.0322 |
temp.reagent | -2.7725 | 0.66 | -4.201 | 0.0059 | 0.0529 |
time | 2.5675 | 0.66 | 3.890 | 0.0079 | 0.0724 |
temp.solvent | 2.3575 | 0.66 | 3.572 | 0.0110 | 0.1016 |
temp.time | -2.3575 | 0.66 | -3.572 | 0.0110 | 0.1016 |
solvent | -2.2175 | 0.66 | -3.360 | 0.0138 | 0.1241 |
time.reagent | -0.6450 | 0.66 | -0.977 | 0.3057 | 0.9955 |
solvent.reagent | 0.4900 | 0.66 | 0.742 | 0.4306 | 1.0000 |
time.solvent | 0.4400 | 0.66 | 0.667 | 0.5393 | 1.0000 |
temp.time.solvent | 0.2450 | 0.66 | 0.371 | 0.7299 | 1.0000 |
time.solvent.reagent | -0.2375 | 0.66 | -0.360 | 0.7384 | 1.0000 |
temp.time.reagent | 0.1950 | 0.66 | 0.295 | 0.7827 | 1.0000 |
temp.time.solvent.reagent | 0.1925 | 0.66 | 0.292 | 0.7849 | 1.0000 |
temp.solvent.reagent | -0.0300 | 0.66 | -0.045 | 0.9661 | 1.0000 |
The function eff.test calculates unadjusted p-values ( p.value ) and simultaneous p-values ( simult.pval ) adjusted to account for multiple testing. Using the latter, from Table 4.5 we see that the main effects of temp and reagent are significant at the experiment-wise 5% level and, obeying effect heredity , their interaction (the p-value is borderline, and hovers around 0.05 depending on simulation error).
The package unrepx also provides the function hnplot to display these results graphically by adding a reference line to a half-normal plot; see Figure 4.5 . The ME and SME lines indicate the absolute size of effects that would be required to reject \(H_0: \theta_i = 0\) at an individual or experimentwise \(100\alpha\) % level, respectively.
Figure 4.5: Desilylation experiment: half-normal plot with reference lines from Lenth’s method.
Informally, factorial effects with estimates greater than SME are thought highly likely to be significant, and effects between ME and SME are considered somewhat likely to be significant (and still worthy of further investigation if the budget allows).
We have identified \(d = 2^f-1\) factorial effects that we wish to estimate from our experiment. As \(d < t = 2^f\) , we can estimate these factorial effects using a full-rank linear regression model.
Let \(t\times d\) matrix \(C\) hold each factorial contrast as a column. Then
\[ \hat{\boldsymbol{\theta}} = C^{\mathrm{T}}\bar{\boldsymbol{y}}\,, \]
with \(\hat{\boldsymbol{\theta}}^{\mathrm{T}} = (\hat{\theta}_1, \ldots, \hat{\theta}_d)\) being the vector of estimated factorial effects and \(\bar{\boldsymbol{y}}^{\mathrm{T}} = (\bar{y}_{1.}, \ldots, \bar{y}_{t.})\) being the vector of treatment means.
We can define an \(n\times d\) expanded contrast matrix as \(\tilde{C} = C \otimes \boldsymbol{1}_r\) , where each row of \(\tilde{C}\) gives the contrast coefficients for each run of the experiment. Then,
\[ \hat{\boldsymbol{\theta}} = \frac{1}{r}\tilde{C}^{\mathrm{T}}\boldsymbol{y}\,. \] To illustrate, we will imagine a hypothetical version of Example 4.1 where each treatment was repeated three times (with \(y_{i1} = y_{i2} = y_{i3}\) ).
If we define a model matrix \(X = \frac{2^{f}}{2}\tilde{C}\) , then \(X\) is a \(n\times d\) matrix with entries \(\pm 1\) and columns equal to unscaled factorial contrasts. Then
\[\begin{align} \left(X^{\mathrm{T}}X\right)^{-1}X^{\mathrm{T}}\boldsymbol{y}& = \frac{1}{n} \times \frac{2^f}{2}\tilde{C}^{\mathrm{T}}\boldsymbol{y}\tag{4.4}\\ & = \frac{1}{2r}\tilde{C}^{\mathrm{T}}\boldsymbol{y}\\ & = \frac{1}{2}\hat{\boldsymbol{\theta}}\,. \\ \end{align}\]
The left-hand side of equation (4.4) is the least squares estimator \(\hat{\boldsymbol{\beta}}\) from the model
\[ \boldsymbol{y}= \boldsymbol{1}_n\beta_0 + X\boldsymbol{\beta} + \boldsymbol{\varepsilon}\,, \] where \(\boldsymbol{y}\) is the response vector and \(\boldsymbol{\varepsilon}\) the error vector from unit-treatment model (4.1) . We have simply re-expressed the mean response as \(\mu + \tau_i = \beta_0 + \boldsymbol{x}_i^{\mathrm{T}}\boldsymbol{\beta}\) , where \(d\) -vector \(\boldsymbol{x}_i\) holds the unscaled contrast coefficients for the main effects and interactions.
We can illustrate these connections for Example 4.1 .
The more usual way to think about this modelling approach is as a regression model with \(f\) (quantitative 29 ) variables, labelled \(x_1, \ldots, x_{2^f-1}\) , scaled to lie in the interval \([-1, 1]\) (in fact, they just take values \(\pm 1\) ). We can then fit a regression model in these variables, and include products of these variables to represent interactions. We usually also include the intercept term. For Example 4.1 :
x | |
---|---|
temp | 8.1200 |
time | 2.5675 |
solvent | -2.2175 |
reagent | 3.0875 |
temp:time | -2.3575 |
temp:solvent | 2.3575 |
temp:reagent | -2.7725 |
time:solvent | 0.4400 |
time:reagent | -0.6450 |
solvent:reagent | 0.4900 |
temp:time:solvent | 0.2450 |
temp:time:reagent | 0.1950 |
temp:solvent:reagent | -0.0300 |
time:solvent:reagent | -0.2375 |
temp:time:solvent:reagent | 0.1925 |
A regression modelling approach is usually more straightforward to apply than defining contrasts in the unit-treatment model, and makes clearer the connection between interaction contrasts and products of main effect contrasts (automatically defined in a regression model). It also enables us to make use of the effects package in R to quickly produce main effect and interaction plots.
Figure 4.6: Desilylation experiment: interaction plot generated using the effects package.
The basic ANOVA table has the following form.
Source | Degress of Freedom | (Sequential) Sum of Squares | Mean Square |
---|---|---|---|
Regression | \(2^f-1\) | \(\sum_{j=1}^{2^f-1}n\hat{\beta}_j^2 - n\bar{y}^2\) | Reg SS/\((2^f-1)\) |
Residual | \(2^f(r-1)\) | \((\boldsymbol{Y}-X\hat{\boldsymbol{\beta}})^{\textrm{T}}(\boldsymbol{Y}-X\hat{\boldsymbol{\beta}})\) | RSS/\((2^f(r-1))\) |
Total | \(2^fr-1\) | \(\boldsymbol{Y}^{\textrm{T}}\boldsymbol{Y}-n\bar{Y}^{2}\) |
The regression sum of squares for a factorial experiment has a very simple form. If we include an intercept column in \(X\) , from Section 1.5.1 ,
\[\begin{align*} \mbox{Regression SS} & = \mbox{RSS(null)} - \mbox{RSS} \\ & = \hat{\boldsymbol{\beta}}^{\mathrm{T}}X^{\mathrm{T}}X\hat{\boldsymbol{\beta}} - n\bar{y}^2 \\ & = \sum_{j=1}^{2^f-1}n\hat{\beta}_j^2 - n\bar{y}^2\,, \end{align*}\]
as \(X^{\mathrm{T}}X = nI_{2^f}\) . Hence, the \(j\) th factorial effect contributes \(n\hat{\beta}_j^2\) to the regression sum if squares, and this quantity can be used to construct a test statistic if \(r>1\) and hence an estimate of \(\sigma^2\) is available. For Example 4.1 , the regression sum of squares and ANOVA table are given in Tables 4.8 and 4.9 .
Sum Sq. | |
---|---|
Regression | 427.2837 |
temp | 263.7376 |
time | 26.3682 |
solvent | 19.6692 |
reagent | 38.1306 |
temp:time | 22.2312 |
temp:solvent | 22.2312 |
temp:reagent | 30.7470 |
time:solvent | 0.7744 |
time:reagent | 1.6641 |
solvent:reagent | 0.9604 |
temp:time:solvent | 0.2401 |
temp:time:reagent | 0.1521 |
temp:solvent:reagent | 0.0036 |
time:solvent:reagent | 0.2256 |
temp:time:solvent:reagent | 0.1482 |
Df | Sum Sq | |
---|---|---|
temp | 1 | 263.7376 |
time | 1 | 26.3682 |
solvent | 1 | 19.6692 |
reagent | 1 | 38.1306 |
temp:time | 1 | 22.2312 |
temp:solvent | 1 | 22.2312 |
temp:reagent | 1 | 30.7470 |
time:solvent | 1 | 0.7744 |
time:reagent | 1 | 1.6641 |
solvent:reagent | 1 | 0.9604 |
temp:time:solvent | 1 | 0.2401 |
temp:time:reagent | 1 | 0.1521 |
temp:solvent:reagent | 1 | 0.0036 |
time:solvent:reagent | 1 | 0.2256 |
temp:time:solvent:reagent | 1 | 0.1482 |
Residuals | 0 | 0.0000 |
A reactor experiment that was presented by Box, Hunter and Hunter (2005, pp259-261) that used a full factorial design for \(m=5\) factors, each at two levels, to investigate the effect of feed rate (litres/min), catalyst (%), agitation rate (rpm), temperature (C) and concentration (%) on the percentage reacted . The levels of the experimental factors will be coded as \(-1\) for low level, and \(1\) for high level. Table 4.10 presents the true factor settings corresponding to these coded values.
Factor | Low level (\(-1\)) | High level (\(1\)) |
---|---|---|
Feed Rate (litres/min) | 10 | 15 |
Catalyst (%) | 1 | 2 |
Agitation Rate (rpm) | 100 | 120 |
Temperature (C) | 140 | 180 |
Concentration (%) | 3 | 6 |
The data from this experiment is given in Table 4.11 .
FR | Cat | AR | Temp | Conc | pre.react |
---|---|---|---|---|---|
-1 | -1 | -1 | -1 | -1 | 61 |
1 | -1 | -1 | -1 | -1 | 53 |
-1 | 1 | -1 | -1 | -1 | 63 |
1 | 1 | -1 | -1 | -1 | 61 |
-1 | -1 | 1 | -1 | -1 | 53 |
1 | -1 | 1 | -1 | -1 | 56 |
-1 | 1 | 1 | -1 | -1 | 54 |
1 | 1 | 1 | -1 | -1 | 61 |
-1 | -1 | -1 | 1 | -1 | 69 |
1 | -1 | -1 | 1 | -1 | 61 |
-1 | 1 | -1 | 1 | -1 | 94 |
1 | 1 | -1 | 1 | -1 | 93 |
-1 | -1 | 1 | 1 | -1 | 66 |
1 | -1 | 1 | 1 | -1 | 60 |
-1 | 1 | 1 | 1 | -1 | 95 |
1 | 1 | 1 | 1 | -1 | 98 |
-1 | -1 | -1 | -1 | 1 | 56 |
1 | -1 | -1 | -1 | 1 | 63 |
-1 | 1 | -1 | -1 | 1 | 70 |
1 | 1 | -1 | -1 | 1 | 65 |
-1 | -1 | 1 | -1 | 1 | 59 |
1 | -1 | 1 | -1 | 1 | 55 |
-1 | 1 | 1 | -1 | 1 | 67 |
1 | 1 | 1 | -1 | 1 | 65 |
-1 | -1 | -1 | 1 | 1 | 44 |
1 | -1 | -1 | 1 | 1 | 45 |
-1 | 1 | -1 | 1 | 1 | 78 |
1 | 1 | -1 | 1 | 1 | 77 |
-1 | -1 | 1 | 1 | 1 | 49 |
1 | -1 | 1 | 1 | 1 | 42 |
-1 | 1 | 1 | 1 | 1 | 81 |
1 | 1 | 1 | 1 | 1 | 82 |
Estimate all the factorial effects from this experiment, and use a half-normal plot and Lenth’s method to decide which are significantly different from zero.
Use the effects package to produce main effect and/or interaction plots for each significant factorial effect from part a.
Now fit a regression model that only includes terms corresponding to main effects and two-factor interactions. How many degrees of freedom does this model use? What does this mean for the estimation of \(\sigma^2\) ? How does the estimate of \(\sigma^2\) from this model relate to your analysis in part a?
x | |
---|---|
FR | -1.375 |
Cat | 19.500 |
AR | -0.625 |
Temp | 10.750 |
Conc | -6.250 |
FR:Cat | 1.375 |
FR:AR | 0.750 |
FR:Temp | -0.875 |
FR:Conc | 0.125 |
Cat:AR | 0.875 |
Cat:Temp | 13.250 |
Cat:Conc | 2.000 |
AR:Temp | 2.125 |
AR:Conc | 0.875 |
Temp:Conc | -11.000 |
FR:Cat:AR | 1.500 |
FR:Cat:Temp | 1.375 |
FR:Cat:Conc | -1.875 |
FR:AR:Temp | -0.750 |
FR:AR:Conc | -2.500 |
FR:Temp:Conc | 0.625 |
Cat:AR:Temp | 1.125 |
Cat:AR:Conc | 0.125 |
Cat:Temp:Conc | -0.250 |
AR:Temp:Conc | 0.125 |
FR:Cat:AR:Temp | 0.000 |
FR:Cat:AR:Conc | 1.500 |
FR:Cat:Temp:Conc | 0.625 |
FR:AR:Temp:Conc | 1.000 |
Cat:AR:Temp:Conc | -0.625 |
FR:Cat:AR:Temp:Conc | -0.500 |
There are several large factorial effects, including the main effects of Catalyst and Temperature and the interaction between these factors, and the interaction between Concentration and Temperature. We can assess their significance using a half-normal plot and Lenth’s method.
We see that PSE = 1.3125, giving individual and simultaneous margins of error of 2.7048 and 5.0625, respectively (where the latter is adjusted for multiple testing). There is a very clear distinction between the five effects which are largest in absolute value and the other factorial effects, which form a very clear line. The five of the largest effects are given in Table 4.13 , are all greater than both margins of error and can be declared as significant.
x | |
---|---|
Cat | 19.50 |
Temp | 10.75 |
Conc | -6.25 |
Cat:Temp | 13.25 |
Temp:Conc | -11.00 |
Figure 4.7: Reactor experiment: interaction plots.
Notice that changing the level of Temperature changes substantial the effect of both Catalyst and Concentration on the response; in particular, the effect of Concentration changes sign depending on the level of Temperature.
This model includes regression parameters corresponding to \(5 + {5 \choose 2} = 15\) factorial effects, plus the intercept, and hence uses 16 degrees of freedom. The remaining 16 degrees of freedom, which were previously used to estimate three-factor and higher interactions, is now used to estimate \(\sigma^2\) , the background variation.
The residual mean square in the reduced model, used to estimate \(\sigma^2\) , is the sum of the sums of squares for the higher-order interactions in the original model, divided by 16 (the remaining degrees of freedom).
(Adapted from Morris, 2011 ) Consider an unreplicated ( \(r=1\) ) \(2^6\) factorial experiment. The total sums of squares,
\[ \mbox{Total SS} = \sum_{i=1}^n(y_i - \bar{y})^2\,, \]
has value 2856. Using Lenth’s method, an informal analysis of the data suggests that there are only three important factorial effects, with least squares estimates
If a linear model including only an intercept and these three effects is fitted to the data, what is the value of the residual sum of squares?
The residual sum of squares has the form
\[ \mbox{RSS} = (\boldsymbol{y}- X\hat {\boldsymbol{\beta}})^{\mathrm{T}}(\boldsymbol{y}- X\hat {\boldsymbol{\beta}})\,, \]
where in this case \(X\) is a \(2^6\times 4\) model matrix, with columsn corresponding to the intercept, main effect of factor \(A\) , the interaction between factors \(A\) and \(B\) , the interaction between factors \(A\) , \(B\) and \(C\) . We can rewrite the RSS as
\[\begin{equation*} \begin{split} \mbox{RSS} & = (\boldsymbol{y}- X\hat {\boldsymbol{\beta}})^{\mathrm{T}}(\boldsymbol{y}- X\hat {\boldsymbol{\beta}}) \\ & = \boldsymbol{y}^{\mathrm{T}}\boldsymbol{y}- 2\boldsymbol{y}^{\mathrm{T}}X\hat {\boldsymbol{\beta}} + \hat {\boldsymbol{\beta}}^{\mathrm{T}}X^{\mathrm{T}}X\hat {\boldsymbol{\beta}} \\ & = \boldsymbol{y}^{\mathrm{T}}\boldsymbol{y}- 2\hat {\boldsymbol{\beta}}^{\mathrm{T}}X^{\mathrm{T}}X\hat {\boldsymbol{\beta}} + \hat {\boldsymbol{\beta}}^{\mathrm{T}}X^{\mathrm{T}}X\hat {\boldsymbol{\beta}} \\ & = \boldsymbol{y}^{\mathrm{T}}\boldsymbol{y}- \hat {\boldsymbol{\beta}}^{\mathrm{T}}X^{\mathrm{T}}X\hat {\boldsymbol{\beta}}\,, \end{split} \end{equation*}\]
as \(\boldsymbol{y}^{\mathrm{T}}X = \hat{\boldsymbol{\beta}}^{\mathrm{T}}X^{\mathrm{T}}X\) .
Due the matrix \(X\) having orthogonal columns, \(X^{\mathrm{T}}X = 2^fI_{p+1}\) , for a model containing coefficients corresponding to \(p\) factorial effects; here, \(p=3\) . Hence,
\[ \mbox{RSS} = \boldsymbol{y}^{\mathrm{T}}\boldsymbol{y}- 2^f \sum_{i=0}^{p}\hat{\beta}_i^2\,. \]
Finally, the estimate of the intercept takes the form \(\hat{\beta}_0 = \bar{Y}\) , and so
\[\begin{equation*} \begin{split} \mbox{RSS} & = \boldsymbol{y}^{\mathrm{T}}\boldsymbol{y}- 2^f\bar{y}^2 - 2^f\sum_{i=1}^{p}\hat{\beta}_i^2 \\ & = \sum_{i=1}^{2^f}(y_i - \bar{y})^2 - 2^f\sum_{i=1}^{p}\hat{\beta}_i^2 \\ & = \mbox{Total SS} - 2^f\sum_{i=1}^{p}\hat{\beta}_i^2\, \end{split} \end{equation*}\]
Recalling that each regression coefficient is one-half of the corresponding factorial effect, for this example we have:
\[ \mbox{RSS} = 2856 - 2^6(1.5^2 + 2^2 + 1^2) = 2392\,. \]
(Adapted from Morris, 2011 ) Consider a \(2^7\) experiment with each treatment applied to two units ( \(r=2\) ). Assume a linear regression model will be fitted containing terms corresponding to all factorial effects.
What is the variance of the estimator of each factorial effect, up to a constant factor \(\sigma^2\) ?
What is the variance of the least squares estimator of \(E(y_{11})\) , the expected value of an observation with the first treatment applied? You can assume the treatments are given in standard order, so the first treatment is defined by setting all factors to their low level. [The answer is, obviously, the same for \(E(y_{12})\) ]. In a practical experimental setting, why is this not a useful quantity to estimate?
What is the variance of the least squares estimator of \(E(y_{11}) - E(y_{21})\) ? You may assume that the second treatment has all factors set to their low levels except for the seventh factor.
Each factorial contrast is scaled so the variance for the estimator is equal to \(4\sigma^2/n = \sigma^2 / 64\) .
\(E(y_{11}) = \boldsymbol{x}_1^{\mathrm{T}}\boldsymbol{\beta}\) , where \(\boldsymbol{x}_1^{\mathrm{T}}\) is the row of the \(X\) matrix corresponding to the first treatment and \(\boldsymbol{\beta}\) are the regression coefficients. The estimator is given by
\[ \hat{E}(y_{11}) = \boldsymbol{x}_1^{\mathrm{T}}\hat{\boldsymbol{\beta}}\,, \]
with variance
\[\begin{align*} \mathrm{var}\left\{\hat{E}(y_{11})\right\} & = \mathrm{var}\left\{\boldsymbol{x}_1^{\mathrm{T}}\hat{\boldsymbol{\beta}}\right\} \\ & = \boldsymbol{x}_1^{\mathrm{T}}\mbox{var}(\hat{\boldsymbol{\beta}})\boldsymbol{x}_1 \\ & = \boldsymbol{x}_1^{\mathrm{T}}\left(X^\mathrm{T}X\right)^{-1}\boldsymbol{x}_1\sigma^2 \\ & = \frac{\boldsymbol{x}_1^{\mathrm{T}}\boldsymbol{x}_1\sigma^2}{2^8} \\ & = \frac{2^7\sigma^2}{2^8} \\ & = \sigma^2 / 2\,. \end{align*}\]
This holds for the expected response from any treatment, as \(\boldsymbol{x}_j^{\mathrm{T}}\boldsymbol{x}_j = 2^7\) for all treatments, as each entry of \(\boldsymbol{x}_j\) is equal to \(\pm 1\) .
This would not be a useful quantity to estimate in a practical experiment, as it is not a contrast in the treatments. In particular, it depends on the estimate of the overall mean, \(\mu\) or \(\beta_0\) (in the unit-treatment or regression model) that will vary from experiment to experiment.
The expected values of \(y_{11}\) and \(y_{21}\) will only differ in terms involving the seventh factor, which is equal to its low level (-1) for the first treatment and its high level (+1) for the second treatment; all the other terms will cancel. Hence
\[ E(y_{11}) - E(y_{21}) = -2\left(\beta_7 + \sum_{j=1}^6\beta_{j7} + \sum_{j=1}^6\sum_{k=j+1}^6\beta_{jk7} + \ldots + \beta_{1234567}\right)\,. \]
The variance of the estimator has the form
\[\begin{align*} \mathrm{var}\left\{\widehat{E(y_{11}) - E(y_{21})}\right\} & = 4\times\mathrm{var}\bigg(\hat{\beta}_7 + \sum_{j=1}^6\hat{\beta}_{j7} + \sum_{j=1}^6\sum_{k=j+1}^6\hat{\beta}_{jk7} + \\ & \ldots + \hat{\beta}_{1234567}\bigg) \\ & = \frac{4\sigma^2}{2\times 2^7}\sum_{j=0}^6{6 \choose j} \\ & = \frac{\sigma^2}{2^6}\times 64 \\ & = \sigma^2\,. \end{align*}\]
Or, as this is a treatment comparison in a CRD, we have
\[ \hat{E}(y_{11}) - \hat{E}(y_{21}) = \widehat{\boldsymbol{c}^{\mathrm{T}}\boldsymbol{\tau}}\,, \]
where \(\boldsymbol{c}\) corresponds to a pairwise treatment comparison, and hence has one entry equal to +1 and one entry equal to -1. From Section 2.5 ,
\[\begin{align*} \mathrm{var}\left(\widehat{\boldsymbol{c}^{\mathrm{T}}\boldsymbol{\tau}}\right) & = \sum_{i=1}^tc_i^2\mathrm{var}(\bar{y}_{i.}) \\ & = \sigma^2\sum_{i=1}^tc_i^2/n_i\,, \end{align*}\]
where in this example \(n_i = 2\) for all \(i\) and \(\sum_{i=1}^tc_i^2 = 2\) . Hence, the variance is again equal to \(\sigma^2\) .
Desilylation is a process of removing silyl, SiH \(_3\) a silicon hydride, from a compound. ↩︎
For two matrices \(A\) and \(B\) of the same dimension \(m\times n\) , the Hadamard product \(A\odot B\) is a matrix of the same dimension with elements given by the elementwise product, \((A\odot B)_{ij} = A_{ij}B_{ij}\) . ↩︎
The absolute value of a normally distributed random variable follows a half-normal distribution. ↩︎
Essentially, \(s_0\) tends in probability to \(\sigma\) as the number of factorial effects tends to infinity. ↩︎
Under \(H_0\) , the \(\hat{\theta}_i\) come from a mean-zero normal distribution, and about 1% of deviates fall outside \(\pm 2.57\sigma^2\) . ↩︎
When qualitative factors only have two levels, each regression term only has 1 degree of freedom, and so there is practically little difference from a quantitative variable. ↩︎
Chapter 9 fractional factorial designs, 9.1 introduction.
Factorial treatment designs are necessary for estimating factor interactions and offer additional advantages (Chapter 6 ). However, their implementation is challenging if we consider many factors or factors with many levels, because the number of treatments then might require prohibitive experiment sizes. Large factorial experiments also pose problems for blocking, since reasonable block sizes that ensure homogeneity of the experimental material within a block are often smaller than the number of treatment level combinations.
For example, a factorial treatment structure with five factors of two levels each already has \(2^5=32\) treatment combinations. An experiment with 32 experimental units then has no residual degrees of freedom, but two full replicates of this design already require 64 experimental units. If each factor has three levels, the number of treatment combinations increases drastically to \(3^5=243\) .
On the other hand, we can often justify the assumption of effect sparsity : effect sizes of high-order interactions are often negligible, especially if interactions of lower orders already have small effect sizes. The key observation for reducing the experiment size is that a large portion of model parameters relate to higher-order interactions: in our example, there are 32 model parameters: one grand mean, five main effects, ten two-way interactions, ten three-way interactions, five four-way interactions, and one five-way interaction. The number of higher-order interactions and their parameters grows fast with increasing number of factors as shown in Table 9.1 for factorials with two factor levels and 3 to 7 factors.
If we ignore three-way and higher interactions in the example, we remove 16 parameters from the model equation and only require 16 observations for estimating the remaining model parameters; this is known as a half-fraction of the \(2^5\) -factorial. Of course, the ignored interactions do not simply vanish, but their effects are now confounded with those of lower-order interactions or main effects. The question then arises: which 16 out of the 32 possible treatment combinations should we consider such that no effect of interest is confounded with a another non-negligible effect?
Factorial | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
---|---|---|---|---|---|---|---|---|
3 | 1 | 3 | 3 | 1 | ||||
4 | 1 | 4 | 6 | 4 | 1 | |||
5 | 1 | 5 | 10 | 10 | 5 | 1 | ||
6 | 1 | 6 | 15 | 20 | 15 | 6 | 1 | |
7 | 1 | 7 | 21 | 35 | 35 | 21 | 7 | 1 |
In this chapter, we discuss the general construction and analysis of fractional replications of \(2^k\) -factorial designs where all factors have two levels. This restriction is often sufficient for practical experiments with many factors, where interest focuses on identifying relevant factors and low-order interactions. We first consider generic factors which we call A , B and so forth, and denote their levels as low (or \(-1\) ) and high (or \(+1\) ). Similar techniques to those discussed here are available for factorials with more than two factors levels and for combination of factors with different number of levels, but the required mathematics is beyond our scope.
We further extend our ideas of fractional replication to deliberately confound some effects with blocks. This allows us to run a \(2^5\) -factorial in blocks of size 16, for example. By altering the confounding between pairs of blocks, we can still recover all effects, albeit with reduced precision.
9.2.1 introduction.
We begin our discussion with the simple example of a \(2^3\) -factorial treatment structure in a completely randomized design. We denote the treatment factors A , B , and C and their levels as \(A\) , \(B\) , and \(C\) with values \(-1\) and \(+1\) . Recall that for any \(2^k\) -factorial, all main effects and all interaction factors (of any order) have one degree of freedom. We can thus also encode the two independent levels of any interaction as \(-1\) and \(+1\) , and we define the level by multiplying the levels of the constituent factors: for \(A=-1\) , \(B=+1\) , \(C=-1\) , the level of A:B is \(AB=A\cdot B=-1\) and the level of A:B:C is \(ABC=A\cdot B\cdot C=+1\) .
It is also convenient to use an additional shorthand notation for a treatment combination, where we use a character string containing the lower-case letter of a treatment factor if it is present on its high level, and no letter if it is present on its low level. For example, we write \(abc\) if A , B , C are on level \(+1\) , and all potential other factors are on the low level \(-1\) , and \(ac\) if A and C are on the high level, and B on its low level. We denote a treatment combination with all factors on their low level by \((1)\) . For a \(2^3\) -factorial, the eight different treatments are then \((1)\) , \(a\) , \(b\) , \(c\) , \(ab\) , \(ac\) , \(bc\) , and \(abc\) .
For example, testing compositions for growth media with factors Carbon with levels glucose and fructose , Nitrogen with levels low and high , and Vitamin with levels Mix 1 and Mix 2 leads to a \(2^3\) -factorial with the 8 possible treatment combinations shown in Table 9.2 .
A | B | C | AB | AC | BC | ABC | Shorthand | |||
---|---|---|---|---|---|---|---|---|---|---|
\(-1\) | \(-1\) | \(-1\) | \(+1\) | \(+1\) | \(+1\) | \(-1\) | \((1)\) | |||
\(-1\) | \(-1\) | \(+1\) | \(+1\) | \(-1\) | \(-1\) | \(+1\) | \(c\) | |||
\(-1\) | \(+1\) | \(-1\) | \(-1\) | \(+1\) | \(-1\) | \(+1\) | \(b\) | |||
\(-1\) | \(+1\) | \(+1\) | \(-1\) | \(-1\) | \(+1\) | \(-1\) | \(bc\) | |||
\(+1\) | \(-1\) | \(-1\) | \(-1\) | \(-1\) | \(+1\) | \(+1\) | \(a\) | |||
\(+1\) | \(-1\) | \(+1\) | \(-1\) | \(+1\) | \(-1\) | \(-1\) | \(ac\) | |||
\(+1\) | \(+1\) | \(-1\) | \(+1\) | \(-1\) | \(-1\) | \(-1\) | \(ab\) | |||
\(+1\) | \(+1\) | \(+1\) | \(+1\) | \(+1\) | \(+1\) | \(+1\) | \(abc\) |
In a \(2^k\) -factorial treatment structure, we estimate main effects and interactions as simple contrasts by subtracting the sum of responses of all observations with the corresponding factors on the low level from those with the factors on the high level. For our example, we estimate the main effect of C-Source (or generically A ) by subtracting all observations with fructose as our carbon source from those with glucose , and averaging: \[\begin{align*} \text{A main effect} &= \frac{1}{4}\left(\,(a-(1)) + (ab-b) + (ac-c) + (abc-bc)\,\right) \\ &= \frac{1}{4}\left(\underbrace{(a+ab+ac+abc)}_{A=+1}-\underbrace{((1)+b+c+bc)}_{A=-1}\right)\;. \end{align*}\] A two-way interaction is a difference of differences and we find the interaction of B with C by first finding the difference between them for A on the low level and for A on the high level: \[ \frac{1}{2}\underbrace{\left((abc-ab)\,-\,(ac-a)\right)}_{A=+1} \quad\text{and}\quad \frac{1}{2}\underbrace{\left((bc-b)\,-\,(c-(1))\right)}_{A=-1}\;. \] The interaction effect is then the averaged difference between the two \[\begin{align*} \text{B:C interaction} &= \frac{1}{4} \left(\;\left((abc-ab)-(ac-a)\right)+\left((bc-b)-(c-(1))\right)\;\right) \\ &= \frac{1}{4} \left(\; \underbrace{(abc+bc+a+(1))}_{BC=+1}\,-\,\underbrace{(ab+ac+b+c)}_{BC=-1}\; \right)\;. \end{align*}\] This value is equivalently found by taking the difference between observations with \(BC=+1\) (the interaction at its ‘high’ level) and \(BC=-1\) (the interaction at its ‘low’ level) and averaging. The other interaction effects are estimated by contrasting the corresponding observations for \(AB=\pm 1\) and \(AC=\pm 1\) , and \(ABC=\pm 1\) , respectively.
We are interested in reducing the size of the experiment and for reasons that will become clear shortly, we choose a design based on measuring the response for four out of the eight treatment combinations. This will only allow estimation of four parameters in the linear model, and exactly which parameters can be estimated depends on the treatments chosen. The question then is: which four treatment combinations should we select?
We investigate three specific choices to get a better understanding of the consequences for effect estimation. The designs are illustrated in Figure 9.1 , where treatment level combinations form a cube with eight vertices, from which four are selected in each case.
Figure 9.1: Some fractions of a \(2^3\) -factorial. A: Arbitrary choice of treatment combinations leads to problems in estimating any effects properly. B: One variable at a time (OVAT) design. C: Keeping one factor at a constant level confounds this factor with the grand mean and creates a \(2^2\) -factorial of the remaining factors.
First, we arbitrarily select the four treatment combinations \((1), a, b, ac\) (Fig. 9.1 A). With this choice, none of the main effects or interaction effects can be estimated using all four data points. For example, an estimate of the A main effect involves \(a-(1)\) , \(ab-b\) , \(ac-c\) , and \(abc-bc\) , but only one of these— \(a-(1)\) —is available in this experiment. Compared to a factorial experiment in four runs, this choice of treatment combinations thus allows using only one-half of the available data for estimating this effect. If we would follow the above logic and contrast the observations with A at the high level with those with A at the low level, thereby using all data, the main effect is estimated as \((ac+a)-(b+(1))\) and obviously leads to a biased and incorrect estimate of the main effect, since the other factors are at ‘incompatible’ levels. Similar problems arise for B and C main effects, where only \(b-(1)\) , respectively \(ac-a\) are available. None of the interactions can be estimated from these data and we are left with a very unsatisfactory muddle of conditional effect estimates that are valid only if other factors are kept at particular levels.
Next, we try to be more systematic and select the four treatment combinations \((1), a, b, c\) (Fig. 9.1 C) where all factors occur on low and high levels. Again, main effect estimates are based on half of the data for each factor, but their calculation is now simpler: \(a-(1)\) , \(b-(1)\) , and \(c-(1)\) , respectively. We note that each estimate involves the same level \((1)\) . This design resembles a one variable at a time experiment, where effects can be estimated individually for each factor, by no estimates of interactions are available. All advantages of a factorial treatment design are then lost.
Finally, we select the four treatment combinations \((1), b, c, bc\) with A on the low level (Fig. 9.1 B). This design is effectively a \(2^2\) -factorial with treatment factors B and C and allows estimation of their main effects and their interaction, but no information is available on any effects involving the third treatment factor A . For example, we estimate the B main effect using \((bc+b)\,-\,(c+(1))\) , and the B:C interaction using \((bc-b)-(c-(1))\) . If we look more closely into Table 9.2 , we find a simple confounding structure: the level of B is always identical to that of A:B . In other words, the two effects are completely confounded in this design, and \((bc+b)\,-\,(c+(1))\) is in fact an estimate of the sum of the B main effect and the A:B interaction. Similarly, C is completely confounded with A:C , and B:C with A:B:C . Finally, the grand mean is confounded with the A main effect; this makes sense since any estimate of the overall average is based only on the low level of A .
Neither of the previous three choices provided a convincing reduction of the factorial design. We now discuss a fourth possibility, the half-replicate of the \(2^3\) -factorial, called a \(2^{3-1}\) -fractional factorial . The main idea is to deliberately alias a high-order interaction with the grand mean. For a \(2^3\) -factorial, we alias the three-way interaction A:B:C by selecting either those four treatment combinations that have \(ABC=-1\) or those that have \(ABC=+1\) . We call the corresponding equation the generator of the fractional factorial; the two possible sets are shown in Figure 9.2 . With either choice, we find three more effect aliases by consulting Table 9.2 . For example, using \(ABC=+1\) as our generator yields the four treatment combinations \(a, b, c, abc\) and we find that A is completely confounded with B:C , B with A:C , and C with A:B .
In this design, any estimate thus corresponds to the sum of two effects. For example, \((a+abc)-(b+c)\) estimates the sum of A and B:C : first, the main effect of A is found as the difference of the runs \(a\) and \(abc\) with A on its high level, and the runs \(b\) and \(c\) with A on its low level: \((a+abc)-(b+c)\) . Second, we contrast runs with B:C on the high level ( \(a\) and \(abc\) ) with those with B:C on its low level ( \(b\) and \(c\) ) for estimating the B:C interaction effect, which is again \((a+abc)-(b+c)\) .
The fractional factorial based on a generator deliberately aliases each main effect with a two-way interaction, and the grand mean with the three-way interaction. This yields a very simple aliasing of effects and each estimate is based on the full data. Moreover, we note that by pooling the treatment combinations over levels of one of the three factors, we create three different \(2^2\) -factorials based on the two remaining factors. For example, ignoring the level of C leads to the full factorial in A and B shown in Figure 9.2 . This is a consequence of the aliasing, as C is completely confounded with A:B .
Figure 9.2: The two half-replicates of a \(2^3\) -factorial with three-way interaction and grand mean confounded. Any projection of the design to two factors yields a full \(2^2\) -factorial design and main effects are confounded with two-way interactions. A: design based on low level of three-way interaction; B: complementary design based on high level.
Our full linear model for a three-factor factorial is \[ y_{ijkl} = \mu + \alpha_i + \beta_j + \gamma_k + (\alpha\beta)_{ij} + (\alpha\gamma)_{ik} + (\beta\gamma)_{jk} + (\alpha\beta\gamma)_{ijk} + e_{ijkl} \] and it contains eight sets of parameters plus the residual variance. In a half-replicate of the \(2^3\) -factorial, we can only estimate the four derived parameters \[ \mu + (\alpha\beta\gamma)_{ijk}, \quad \alpha_i + (\beta\gamma)_{jk}, \quad \beta_j + (\alpha\gamma)_{ik}, \quad \gamma_k + (\alpha\beta)_{ij}\;. \] These provide the alias sets of confounded parameters, where only the sum of parameters in each set can be estimated: \[ \{1, ABC\}, \quad \{A, BC\}, \quad \{B, AC\}, \quad \{C, AB\}\;. \]
If the three interactions are negligible, then our four estimates correspond exactly to the grand mean and the three main effects. This corresponds to an additive model without interactions and allows a simple and clean interpretation of the parameter estimates. For example, with \((\beta\gamma)_{jk}=0\) , the second derived parameter is now identical to \(\alpha_i\) .
It might also be the case that the A and B main effects and their interaction are the true effects, while the factor C plays no role. The estimates of the four derived parameters are now estimates of the parameters \(\mu\) , \(\alpha_i\) , \(\beta_j\) , and \((\alpha\beta)_{ij}\) , while \(\gamma_k=(\alpha\gamma)_{ik}=(\beta\gamma)_{jk}=(\alpha\beta\gamma)_{ijk}=0\) .
Many other combinations are possible, but the aliasing in the \(2^{3-1}\) -fractional factorial does not allow us to distinguish the different interpretations without additional experimentation.
The half-replicate of a \(2^3\) -factorial does not provide an entirely convincing example for the usefulness of fractional factorial designs due to the complete confounding of main effects and two-way interactions, both of which are typically of great interest. With more factors in the treatment structure, however, we are able to alias interactions of higher order and confound low-order interactions of interest with high-order interactions that we might assume negligible.
The generator or generating equation provides a convenient way for constructing fractional factorial designs. The generator is then a word written by concatenating the factor letters, such that \(AB\) denotes a two-way interaction, and our previous example \(ABC\) is a three-way interaction; the special ‘word’ \(1\) denotes the grand mean. A generator is then a formal equation that identifies two words and enforces the equality of the corresponding treatment combinations. In our \(2^{3-1}\) design, the generator \[ ABC=+1\;, \] selects all those rows in Table 9.2 for which the relation is true, i.e., for which \(ABC\) is on the high level.
A generator determines the effect confounding of the experiment: the generator itself is one confounding and \(ABC=+1\) describes the complete confounding of the the three-way interaction A:B:C with the grand mean.
From the generator, we can derive all other confoundings by simple algebraic manipulation. By formally ‘multiplying’ the generator with an arbitrary word, we find a new relation between effects. In this manipulation, the multiplication with the letter \(+1\) leaves the equation unaltered, multiplication with \(-1\) inverses signs, and a product of two identical letters yields \(+1\) . For example, multiplying our generator \(ABC=+1\) with the word \(B\) yields \[ ABC\cdot B=(+1)\cdot B \iff AC=B\;. \] In other words, the B main effect is confounded with the A:C interaction. Similarly, we find \(AB=C\) and \(BC=A\) as two further confounding relations by multiplying the generator with \(C\) and \(A\) , respectively.
Further trials with manipulating the generator show that no further relations can be obtained. For example, multiplying \(ABC=+1\) with the word \(AB\) yields \(C=AB\) again, and multiplying this relation with \(C\) yields \(C\cdot C=AB\cdot C\iff +1=ABC\) , the original generator. This means that indeed, we have fully confounded four pairs of effects and no others. In general, a generator for a \(2^k\) factorial produces \(2^k/2=2^{k-1}\) such alias relations between factors, so we have a direct way to check if we found all. In our example, \(2^3/2=2^2=4\) , so our alias relations \(ABC=+1\) , \(AB=C\) , \(AC=B\) , and \(BC=A\) cover all existing confoundings.
This property also means that by choosing any of the implied relations as our generator, we get exactly the same set of treatment combinations. For example, instead of \(ABC=+1\) , we might equally well choose \(A=BC\) ; this selects the same set of rows and implies the same set of confounding relations. Usually, we use a generator that aliases a high-order interaction with the grand mean, simply because it is the most obvious and convenient thing to do.
Useful fractions of factorial designs with manageable aliasing are associated with a generator, because then can effects be properly estimated and meaningful confounding arises. Each generator selects one-half of the possible treatment combinations and this is the reason why we set out to choose four rows for our examples, and not, say, six.
We briefly note that our first and second choice in Section 9.2.3 are not based on a generator, leaving us with a complex partial confounding of effects. In contrast, our third choice selected all treatments with A on the low level and does have a generator, namely \[ A=-1\;. \] Algebraic manipulation then shows that this design implies the additional three confounding relations \(AB=-C\) , \(AC=-B\) , and \(ABC=-BC\) . In other words, any effect involving the factor A is confounded with another effect not involving that factor, which we easily verify from Table 9.2 .
Generators and their algebraic manipulation provide an efficient way for finding the confoundings in higher-order factorials, where looking at the corresponding table of treatment combinations quickly becomes unfeasible. As we can see from the algebra, the most useful generator is always confounding the grand mean with the highest-order interaction.
For four factors, this generator is \(ABCD=+1\) and we expect that there are \(2^4/2=8\) relations in total. Multiplying with any letter reveals that main effects are then confounded with three-way interactions, such as \(ABCD=+1\iff BCD=A\) after multiplying with \(A\) , and similarly \(B=ACD\) , \(C=ABD\) , and \(D=ABC\) . Moreover, by multiplication with two-letter words we find that all two-way interactions are confounded with other two-way interactions, namely via the three relations \(AB=CD\) , \(AC=BD\) , and \(AD=BC\) . This is already an improvement over fractions of the \(2^3\) -factorial, especially if we can make the argument that three-way interactions can be neglected and we thus have direct estimates of all main effects. If we find a significant and large two-way interaction— A:B , say—then we cannot distinguish if it is A:B , its alias C:D , or a combination of the two that produces the effect. Subject-matter considerations might be available to separate these possibilities. If not, there is at least a clear goal for a subsequent experiment to disentangle the two interaction effects.
Things improve further for five factors and the generator \(ABCDE=+1\) which reduces the number of treatment combinations from \(2^5=32\) to \(2^{5-1}=16\) . Now, main effects are confounded with four-way interactions, and two-way interactions are confounded with three-way interactions. Invoking the principle of effect sparsity and neglecting the three- and four-way interactions yields estimable main effects and two-way interactions.
Starting from factorials with six factors, main effects and two-way interactions are confounded with interactions of order five and four, respectively, which in most cases can be assumed to be negligible.
A simple way for creating the design table of a fractional factorial using R exploits these algebraic manipulations: first, we define our generator. We then create the full design table with \(k\) columns, one for each treatment factor, and one row for each of the \(2^k\) combinations of treatment levels, where each cell is either \(-1\) or \(+1\) . Next, we create a new column for the generator and calculate its entries by multiplying the corresponding columns. Finally, we remove all rows for which the generator equation is not fulfilled and keep the remaining rows as our design table. For a 3-factor design with generator \(ABC=-1\) , we create three columns \(A\) , \(B\) , \(C\) and eight rows. The new column \(ABC\) has entries \(A\cdot B\cdot C\) , and we delete those rows for which \(A\cdot B\cdot C\not=-1\) .
As a larger example of a fractional factorial treatment design, we discuss an experiment conducted during the sequential optimization of a yeast growth medium optimization. The overall aim was to find a medium composition that maximizes growth, and we discuss this aspect in more detail in Chapter 10 . Here, we concentrate on determining the individual and combined effects of five medium ingredients—glucose Glc , two different nitrogen sources N1 (monosodium glutamate) and N2 (an amino acid mixture), and two vitamin sources Vit1 and Vit2 —on the resulting number of yeast cells. Different combinations of concentrations of these ingredients are tested on a 48-well plate, and the growth curve is recorded for each well by measuring the optical density over time. We use the increase in optical density ( \(\Delta\text{OD}\) ) between onset of growth and flattening of the growth curve at the diauxic shift as a rough but sufficient approximation for increase in number of cells.
To determine how the five medium components influence the growth of the yeast culture, we used the composition of a standard medium as a reference point, and simultaneously altered the concentrations of the five components. For this, we selected two concentrations per component, one lower, the other higher than the standard, and considered these as two levels for each of five treatment factors. The treatment structure is then a \(2^5\) -factorial and would in principle allow estimation of the main effects and all two-, three-, four-, and five-factor interactions when all \(32\) possible combinations are used. However, a single replicate would require two-thirds of a plate and this is undesirable because we would like sufficient replication and also be able to compare several yeast strains in the same plate. Both requirements can be accommodated by using a half-replicate of the \(2^5\) -factorial with 16 treatment combinations, such that three independent experiments fit on a single plate.
A generator \(ABCDE=1\) confounds the main effects with four-way interactions, which we consider negligible for this experiment. Still, two-way interactions are confounded with three-way interactions, and in the first implementation we assume that three-way interactions are much smaller than two-way interactions. We can then interpret main effect estimates directly, and assume that derived parameters involving two-way interactions have only small contributions from the corresponding three-way interactions.
A single replicate of this \(2^{5-1}\) -fractional factorial generates 16 observations, sufficient for estimating the grand mean, five main effects, and the ten two-way interactions, but we are left with no degrees of freedom for estimating the residual variance. We say the design is saturated . This problem is circumvented by using two replicates of this design per plate. While this requires 32 wells, the same size as the full factorial, this strategy produces duplicate measurements of the same treatment combinations which we can manually inspect for detecting errors and aberrant observations. The 16 treatment combinations considered are shown in Table 9.3 together with the measured difference in OD for the first and second replicate, with higher differences indicating higher growth.
Glc | N1 | N2 | Vit1 | Vit2 | Growth_1 | Growth_2 |
---|---|---|---|---|---|---|
20 | 1 | 0 | 1.5 | 4 | 1.7 | 35.68 |
60 | 1 | 0 | 1.5 | 0 | 0.1 | 67.88 |
20 | 3 | 0 | 1.5 | 0 | 1.5 | 27.08 |
60 | 3 | 0 | 1.5 | 4 | 0.0 | 80.12 |
20 | 1 | 2 | 1.5 | 0 | 120.2 | 143.39 |
60 | 1 | 2 | 1.5 | 4 | 140.3 | 116.30 |
20 | 3 | 2 | 1.5 | 4 | 181.0 | 216.65 |
60 | 3 | 2 | 1.5 | 0 | 40.0 | 47.48 |
20 | 1 | 0 | 4.5 | 0 | 5.8 | 41.35 |
60 | 1 | 0 | 4.5 | 4 | 1.4 | 5.70 |
20 | 3 | 0 | 4.5 | 4 | 1.5 | 84.87 |
60 | 3 | 0 | 4.5 | 0 | 0.6 | 8.93 |
20 | 1 | 2 | 4.5 | 4 | 106.4 | 117.48 |
60 | 1 | 2 | 4.5 | 0 | 90.9 | 104.46 |
20 | 3 | 2 | 4.5 | 0 | 129.1 | 157.82 |
60 | 3 | 2 | 4.5 | 4 | 131.5 | 143.33 |
Clearly, the medium composition has a huge impact on the resulting growth, ranging from a minimum of 0 to a maximum of 181. The original medium has an average ‘growth’ of \(\Delta\text{OD}\approx 80\) , and this experiment already reveals a condition with approximately 2.3 fold increase. We also see that measurement with N2 at the low level are abnormally low in the first replicate. We remove these eight values from our analysis. 13
Our fractional factorial design has five treatment factors and several interaction factors, and we use an analysis of variance initially to determine which of the medium components has an appreciable effect on growth, and how the components interact. The full model is growth~Glc*N1*N2*Vit1*Vit2 , but only half of its parameters can be estimated. Since we deliberately confounded effects in our fractional factorial treatment structure, we know which derived parameters are estimated, and can select one member of each alias set for our model. The model specification growth~(Glc+N1+N2+Vit1+Vit2)^2 asks for an ANOVA based on all main effects and all two-way interactions (it expands to growth~Glc+N1+N2+...+Glc:N1+...+Vit1:Vit2 ). After pooling the data from both replicates and excluding the aberrant N2 observation of the first replicate, the resulting ANOVA table is
Df | Sum Sq | Mean Sq | F value | Pr(>F) | |
---|---|---|---|---|---|
1 | 6148 | 6148 | 26.49 | 0.0008772 | |
1 | 1038 | 1038 | 4.475 | 0.0673 | |
1 | 34298 | 34298 | 147.8 | 1.94e-06 | |
1 | 369.9 | 369.9 | 1.594 | 0.2423 | |
1 | 6040 | 6040 | 26.03 | 0.0009276 | |
1 | 3907 | 3907 | 16.84 | 0.003422 | |
1 | 1939 | 1939 | 8.357 | 0.02017 | |
1 | 264.8 | 264.8 | 1.141 | 0.3166 | |
1 | 753.3 | 753.3 | 3.247 | 0.1092 | |
1 | 0.9298 | 0.9298 | 0.004007 | 0.9511 | |
1 | 1450 | 1450 | 6.248 | 0.03697 | |
1 | 9358 | 9358 | 40.33 | 0.0002204 | |
1 | 277.9 | 277.9 | 1.198 | 0.3057 | |
1 | 811.4 | 811.4 | 3.497 | 0.0984 | |
1 | 1280 | 1280 | 5.515 | 0.0468 | |
8 | 1856 | 232 |
We find several substantial main effects in this analysis, with N2 the main contributor followed by Glc and Vit2 . Even though N1 has no significant main effect, it appears in several significant interactions; this also holds to a lesser degree for Vit1 . Several pronounced interactions demonstrate that optimizing individual components will not be a fruitful strategy, and we need to simultaneously change multiple factors to maximize the growth. This information can only be acquired by using a factorial design.
We do not discuss the necessary subsequent analyses of contrasts and effect sizes for the sake of brevity; they work exactly as for smaller factorial designs.
Since the design is saturated, a single replicate does not provide information about uncertainty. If only the single replicate can be analyzed, we have to reduce the model to free up degrees of freedom from parameter estimation to estimate the residual variance. If subject-matter knowledge is available to decide which factors can be safely removed without missing important effects, then a single replicate can be a successfully analysed. For example, knowing that the two nitrogen sources and the two vitamin components do not interact, we might specify the model Growth~(Glc+N1+N2+Vit1+Vit2)^2 - N1:N2 - Vit1:Vit2 that removes the two corresponding interactions while keeping the three remaining ones. This strategy is somewhat unsatisfactory, since we now still only have two residual degrees of freedom and correspondingly low precision and power, and we cannot test if removal of the factors was really justified. Without good subject-matter knowledge, this strategy can give very misleading results if significant and large effects are removed from the analysis.
The definition of a single generator creates a half-replicate of the factorial design. For higher-order factorials starting with the \(2^5\) -factorials, useful designs are also available for higher fractions, such as quarter-replicates that would require only 8 of the 32 treatment combinations in a \(2^5\) -factorial. These designs are constructed by using more than one generator, which also leads to more complicated confounding.
For example, a quarter-fractional requires two generators: one generator to specify one-half of the treatment combinations, and a second generator to specify one-half of those. Both generators introduce their own aliases which we determine using the generator algebra. In addition, multiplying the two generators introduces further aliases through the generalized interaction .
As a first example, we construct a quarter-replicate of a \(2^5\) -factorial. Which two generators should we use? Our first idea is probably to use the five-way interaction for defining the first set of aliases, and one of the four-way interactions for defining the second set. We might choose the two generators \(G_1\) and \(G_2\) as \[ G_1: ABCDE=1 \quad\text{and}\quad G_2: BCDE=1\;, \] for example. The resulting eight treatment combinations are shown in Table 9.4 (left). We see that in addition to the two generators, we also have a further highly undesirable confounding of the main effect of A with the grand mean: the column \(A\) only contains the high level. This is a consequence of the interplay of the two generators, and we find this additional confounding directly by comparing the left- and right-hand side of their generalized interaction: \[ G_1G_2 = ABCDE\cdot BCDE=ABBCCDDEE = A =1\;. \]
IMAGES
VIDEO
COMMENTS
Factorial experiment
14.2: Design of experiments via factorial designs
Full Factorial Design: Understanding the Impact of ...
Chapter 8 Factorial Experiments
All Topics Factorial Design of Experiments
What is a Factorial Design of an Experiment?
Topic 9. Factorial Experiments [ST&D Chapter 15]
Figure 9.1 Factorial Design Table Representing a 2 × 2 Factorial Design. In principle, factorial designs can include any number of independent variables with any number of levels. For example, an experiment could include the type of psychotherapy (cognitive vs. behavioral), the length of the psychotherapy (2 weeks vs. 2 months), and the sex of ...
This benefit arises from factorial experiments rather than single factor experiments with n observations per cell. An alternative design choice could have been to do two one-way experiments, one with a treatments and the other with b treatments, with n observations per cell. ... There is a similar equation for factor B. \(\phi^{2} = ( na \times ...
A full factorial design (also known as complete factorial design) is the most complete of the design options, meaning that each factor and level are combined to test every possible combination condition. Let us expand upon the theoretical ERAS factorial experiment as an illustrative example. We designed our own ERAS protocol for Whipple procedures, and our objective is to test which components ...
Chapter 4 Factorial experiments. In Chapters 2 and 3, we assumed the objective of the experiment was to investigate \(t\) unstructured treatments, defined only as a collection of distinct entities (drugs, advertisements, receipes, etc.). That is, there was not necessarily any explicit relationship between the treatments (although we could clearly choose which paticular comparisons between ...
Lesson 6: The \(2^k\) Factorial Design
Introduction. Factorial designs are the basis for another important principle besides blocking - examining several factors simultaneously. We will start by looking at just two factors and then generalize to more than two factors. Investigating multiple factors in the same design automatically gives us replication for each of the factors.
ORIAL DESIGNS4.1 Two Factor Factorial DesignsA two-factor factorial design is an experimental design in which data is collected for all possible combination. sible factor combinations then the de. ign is abalanced two-factor factorial design.A balanced a b factorial design is a factorial design for which there are a levels of factor A, b levels ...
Factorial Experiment
Chapter 6 of BHH (2nd ed) discusses fractional factorial designs. Example: full 25 factorial would require 32 runs. An experiment with only 8 runs is a 1/4th (quarter) fraction. Because 1⁄4=(1⁄2)2=2-2, this is referred to as a 25-2 design. In general, 2k-p design is a (1⁄2)p fraction of a 2k design using 2k-p runs.
2k Factorial Models Design of Experiments - Montgomery Chapter 6 23 2k Factorial Design † Each factor has two levels (often labeled + and ¡) † Very useful design for preliminary analysis † Can \weed out" unimportant factors † Also allows initial study of interactions † For general two-factor factorial model y ijk = „ + fi i + fl j +(fifl) ij + † ijk † Have 1+(a ¡ 1) + (b ...
The two-way ANOVA is probably the most popular layout in the Design of Experiments. To begin with, let us define a factorial experiment: An experiment that utilizes every combination of factor levels as treatments is called a factorial experiment. Model for the two-way factorial experiment. In a factorial experiment with factor A at a levels ...
It is easy to see that with the addition of more and more crossed factors, the replicate size will increase rapidly and design modifications have to be made to make the experiment more manageable. In a factorial experiment, as combinations of different factor levels play an important role, it is important to differentiate between the lone (or ...
The generator or generating equation provides a convenient way for constructing fractional factorial designs. The generator is then a word written by concatenating the factor letters, such that \(AB\) denotes a two-way interaction, and our previous example \(ABC\) is a three-way interaction; the special 'word' \(1\) denotes the grand mean.
11.3: Two-Way ANOVA (Factorial Design)
The degrees of freedom (v 1 and v 2) for the F ratio are the degrees of freedom associated with the effects used to compute the F ratio. For example, consider the F ratio for Factor A when Factor A is a fixed effect. That F ratio (F A) is computed from the following formula: F A = F (v 1, v 2) = MS A / MS WG.
The factorial experiment is a multifactor cross-grouping experiment that can test the differences between each factor and examine the interaction effects between factors . Therefore, the interactions between different concentrations of conditioning additives based on the factorial experiment need to be further studied.
5.1 - Factorial Designs with Two Treatment Factors