Sunteți pe pagina 1din 7

Section on Physical and Engineering Sciences

Comparison of Slack Variable and Mixture Experiment Approaches


Samantha M. Landmesser1 and Greg F. Piepel2
1
Department of Statistics, University of Tennessee, Knoxville, TN
2
Statistical Sciences, Pacific Northwest National Laboratory, Richland, WA

Abstract experiments. Using an approach specifically for mixture


experiments provides several advantages over the SV
In a mixture experiment, the response variable depends on approach. Using mixture experiment examples from the
the proportions of the components, which must sum to literature, we illustrate that the mixture approach for
one. Because of this constraint, standard polynomial modeling and data analysis provides results that are the
models cannot be used to analyze mixture experiment same or better without the risk of incorrect conclusions
data. To get around this, some researchers ignore one of that can occur with the SV approach.
the components and use standard polynomial models in
the remaining components. Because the component A SV design involves setting the values of all but one of
proportions must sum to one, the ignored component the components, often using a standard statistical design
(referred to as the slack variable (SV)) makes up the such as a fractional factorial, central composite, or Box-
remaining proportion of the mixture. In the literature, Behnken design. The remaining component, called the
there have been many examples of researchers using the slack variable, has a proportion equal to one minus the
SV approach instead of a mixture approach for modeling. sum of the proportions of the other components. When
We have analyzed data from several of these examples modeling the data from these types of experiments, it is
using both approaches. For examples whose goal was to often assumed that the SV doesn’t have a significant
screen the mixture components (screening examples), we effect on the response. An inactive “filler” component is
fit full linear models and identified which components an example of a component having no effect. Practitioners
were important using both approaches. In the screening also often assume the SV has no effect if it makes up the
examples, the mixture modeling approach revealed that vast majority of the mixture. In some cases, practitioners
the SV had a significant effect on the response. For assume that the SV has at most an additive linear effect. A
examples that had sufficient data to fit quadratic models diluent component (i.e., a component that dilutes the
to the data (quadratic examples), we used stepwise mixture) would be an example of a component with a
regression to develop reduced quadratic models for the linear additive effect.
SV approach, and partial quadratic mixture (PQM)
models for the mixture approach. In the quadratic The mixture experiment approach takes into account the
examples, the PQM models identified the SV and/or one unique restrictions of mixture experiments. Specifically,
of its quadratic blending terms as having a significant the proportions of the components (xi) must fall between
effect on the response variable. Hence, by completely zero and one ( 0 ≤ x i ≤ 1 , i = 1, 2, … , q) and must sum to
ignoring a component’s effect on the response, SV q
analysis carries a greater risk of wrong conclusions than one ( ∑ x i = 1 ). The mixture approach uses every
the mixture approach. There are fewer possible reduced i =1

quadratic SV models than possible PQM models because component in the design of the experiment and the
the reduced quadratic SV models are a subset of the class analysis of the data. This approach does not make any
of PQM models. As a result, PQM models will always fit assumptions about the SV or require any specific situation
the data as well as, or better than, the best reduced to be applicable.
quadratic SV model. We conclude that it is better to
analyze mixture experiments using methods specifically Previous literature contains several examples of mixture
developed for them instead of using standard methods experiments analyzed using the SV approach. Table 1
with the SV approach. briefly summarizes some aspects of several mixture
experiment examples from the literature that used the SV
Keywords: Mixture Components, Model Reduction, approach. In each of the examples, the researchers used
Partial Quadratic Mixture Models, Variable Selection. the SV approach to analyze the data and make
conclusions about how the variables affected the
1. Introduction response. We analyzed data from several of these
examples using the mixture and SV approaches. The
In a mixture experiment, the response variable depends on examples fall into two groups, screening examples and
the proportions of the components, which must sum to quadratic examples. In the screening examples, the goal
one. Many researchers use the slack variable (SV) was to identify the components having significant effects
approach when designing and analyzing mixture on a response variable. In quadratic examples, the goal

1711
Section on Physical and Engineering Sciences

Table 1: Summary of Slack Variable Experiments from the Literature


SV
SV
(a) (a) (a) (b) Linear
Slack Variable (SV) Example q n/d #RVs SV Component SV Range Inactive?
(c) Additive?
(d)

MacDonald and Bly (1966) 5 27/25 8 Shortening (0.030, 0.160) No No


Fonner et al. (1970) 3 9/9 4 Dicalphos (0.0171, 0.9314) No No
Cain and Price (1986) 4 19/19 3 Base Oil (0.845, 0.925) No No
Goh and Roy (1989) 11 16/16 5 SiO2 (0.364, 0.495) No No
Takayama and Nagai (1991) 5 12/9 3 Water (0.4011, 0.7005) No No
Chan & Kavanagh (1992) 8 12/12 7 Water (0.7926, 0.9352) No No
Abdullah et al. (1993) 3 9/9 6 Water (0.3086, 0.5110) No No
Guillou & Floros (1993) 4 15/13 5 Brine Base (0.894, 0.998) No No
Levison et al. (1994) 6 36/27 3 Water (0.355, 0.775) No No
Piepel and Cornell (1994) 4 20/15 4 Water (0.85, 1.00) No No
Setz et al. (1997) 5 24/21 3 TPE (0.120, 0.750) No No
Cornell (2000) Example #1(e) 3 9/7 1 Varied (0, 1.00) No No
Cornell (2000) Example #2 3 13/9 1 Varied (0, 1.00) No No
Cornell (2000) Example #3(e) 4 20/10 1 Varied (0, 1.00) No No
Cornell & Gorman (2003) 4 14/12 1 x4 (0.949, 0.975) No No
(a) q = the number of mixture components that were varied, n = the total number of data points, d = the number of
distinct data points, # RVs = the number of response variables.
(b) The component proportion range of the SV. If the experiment had some components constant, the range shown is
for proportion of the SV component relative to the components that were varied.
(c) This column indicates whether the SV component is inactive for one or more of the response variables.
(d) This column indicates whether the SV component has a linear additive effect on one or more response variables.
(e) This example was also investigated by Khuri (2005).

was to develop the best quadratic model that adequately 2.1 Slack Variable Models
approximates the relationship between the response
variable and the mixture components. The SV approach is sometimes used by practitioners for
screening, that is, discovering which mixture components
Section 2 presents the mixture and SV models used to fit are important for a given response variable. In this case,
data from the literature examples and the summary the full linear SV model is used
statistics used to assess and compare the models. Section q −1
3 presents the results of fitting mixture and SV models for E ( y ) = α 0 + ∑ α i xi (1)
i =1
selected examples. We also discuss some examples for
More often, researchers are interested in obtaining an
which the mixture approach resulted in better models and
adequate approximation of the true relationship between
explain why that occurred. Section 4 summarizes the
the response variable and mixture compositions. In that
work and explains why the mixture approach is more
case, the full quadratic SV model
appropriate in all situations than the SV approach. q −1 q −1 q − 2 q −1
E ( y ) = α 0 + ∑ α i xi + ∑ α ii xi2 + ∑ ∑ α ij xi x j (2)
2. Methods i =1 i =1 i =1 j =i +1
is often used. Removing non-significant terms from the
Both mixture and SV models were used to analyze the full quadratic SV model in Equation (2) results in a
data from the literature examples. For each screening reduced quadratic SV model.
example, we fit the full linear models (mixture and SV)
and identified significant terms. For each quadratic 2.2 Mixture Experiment Models
example, we fit the full quadratic models (mixture and
SV) and then reduced them using stepwise regression. We Mixture models include every component in the model,
determined the best-fitting reduced model based on and don’t have an intercept (constant term). The most
regression summary statistics. Each of the models and commonly used mixture models are the Scheffé models
summary statistics used is discussed in this section.

1712
Section on Physical and Engineering Sciences

(Scheffé 1958, Cornell 2002, Smith 2005). The full linear Piepel, Szychowski, and Loeppky (2002) discuss partial
Scheffé model is quadratic mixture (PQM) models that contain a subset of
q all quadratic (squared and crossproduct) terms including
E ( y ) = ∑ β i xi (3) those involving the SV. A PQM model has the general
i =1
form
When a researcher wants to investigate the quadratic
q
blending effects of mixture components on the response E ( y) = ∑ α i xi
variable, the full quadratic Scheffé model i =1
q q −1 q (6)
E ( y ) = ∑ α i x i + ∑ ∑ β ij x i x j (4)
⎧q 2
q −1 q ⎫
+ Selected ⎨ ∑ β ii x i + ∑ ∑ β ij x i x j ⎬
i =1 i =1 j = i +1
⎩i =1 i =1 j = i +1 ⎭
is appropriate. Notice that linear as well as nonlinear where “Selected” means that only a subset of the squared
blending terms involving the component treated as the SV and crossproduct terms can be chosen. The main
(that is, xq) are included in model (4). This allows limitation is that if a squared term for a component is
researchers to investigate the effects of the SV that are selected, all of the crossproduct terms involving that
ignored in the SV models (1) and (2). component cannot also be selected. Doing so would result
in an exact collinearity in the model. Variable selection
The linear mixture model (3) is not useful for directly methods (such as stepwise regression) would never select
testing the significance of component effects on a terms forming an exact collinearity, so this is not an issue
response variable (see Cornell 2002 and Smith 2005). of practical concern.
Instead, the component slope linear mixture (CSLM)
model discussed by Piepel (2006, 2007) can be used for Note that the class of reduced quadratic SV models and
this purpose. The CSLM model is given by the class of reduced quadratic Scheffé models are both
q q
E( y ) = γ 0 + ∑ γ i (1 − si ) xi = γ 0 + ∑ γ i xi' (5a) subclasses of the class of PQM models. That is, the class
i =1 i =1 of PQM models contains models that are equivalent to all
where xi’ = (1 − si)xi and the coefficients γi (i = 1, 2, … , reduced quadratic SV models and all reduced quadratic
q) are subject to the restriction Scheffé models. The class of PQM models also contains
models not in the classes of reduced quadratic SV models
( )
q
∑γ j 1 − s j s j = 0 (5b) and reduced quadratic Scheffé models. Hence, it is
j =1
guaranteed that the best-fitting PQM model will always fit
with s = (s1, …, sq) being a reference mixture. In (5a), γ0 as well or better than any reduced quadratic SV model or
is the response value at the reference mixture and γi reduced quadratic Scheffé model.
represents the slope of the response surface along the Cox
effect direction (Cornell 2002, Section 5.9) for the ith 2.4 Statistics Used to Evaluate and Compare Models
component. The CSLM model (5) provides the same fit to
data as the linear mixture model (3). Because the γi Three different R2 statistics were used to evaluate mixture
coefficients are the slopes of the response along and SV models. These are denoted as R 2 , R A2 , and R P2 ,
component effect directions, standard t-tests can be used
to assess whether the mixture components have which are defined as follows. The (ordinary) R2 statistic is
significant effects on the response variable. given by
n
2
∑ ( ŷi − yi )
2.3 Reduced Quadratic Models
R 2 = 1 − i =n1 (7)
2
Because we are interested in identifying which quadratic ∑ ( yi − y )
i =1
terms have a significant effect on the response, the full and is interpreted as the fraction of variability in the data
quadratic models [Equations (2) and (4)] are reduced by accounted for by the fitted model. The adjusted R2
removing nonsignificant terms. When the full quadratic statistic is given by
SV model in Equation (2) is reduced to contain a subset n
2
of the quadratic terms not involving the SV, the resulting ∑ ( ŷi − yi ) ( n − p )
i =1
model is called a reduced quadratic SV model. When the R A2 = 1 − n
(8)
full quadratic Scheffé model in Equation (4) is reduced to 2
∑ ( yi − y ) ( n − 1)
contain a subset of the crossproduct terms (including the i =1
crossproduct terms involving the SV), the resulting model and is interpreted as the adjusted fraction of variability in
is called a reduced quadratic mixture model. the data accounted for by the fitted model. The adjustment
is for the number of parameters (p) and number of data

1713
Section on Physical and Engineering Sciences

points (n) used in fitting the model. The predicted R2 mixture components on a response variable of interest
statistic is given by when using a fitted mixture model. The model is used to
n
2 predict, for each component, the response for a series of
∑ ( ŷ(i ) − yi ) compositions lying along an effect direction (e.g., the Cox
RP2 = 1 − i =1n (9) effect direction) for that component. Along such a
2
∑ ( yi − y ) direction, the component of interest is varied within the
i =1
allowable composition region of interest. The remaining
and is calculated by a method equivalent to leaving each components change proportionally to offset the changes
data point out of the model fit, and then evaluating how in the component of interest. The predicted response
well the model predicts the property for that data point. values are plotted on the y-axis and changes in each
R P2 estimates the fraction of variability that would be component from its reference mixture value are plotted on
explained in predicting new observations drawn from the the x-axis. The predicted response values along the effect
same mixture composition space. direction for a given component form a component
response trace. The response traces for the components
The Root Mean Square Error is given by varied in a mixture experiment plotted together form a
component response trace plot.
n
2
∑ ( ŷi − yi )
i =1 Components with steeper response traces have stronger
RMSE = (10)
n− p effects on the response. A response trace that is nearly
horizontal indicates the corresponding component has
If the fitted model is adequate and does not have a
little or no effect on the response. Components whose
statistically significant lack-of-fit, this statistic provides
response traces are very close may have similar effects on
an estimate of the standard deviation representing the
the response. Thus, component response trace plots can be
experimental and measurement uncertainty associated
used to guide the reduction of components in a mixture
with the processes of conducting the experiment and
experiment model (e.g., see Piepel and Redgate 1997).
measuring the response variable.
3. Results
2.5 Component Response Trace Plots
Each literature example listed in Table 1 was analyzed
Component response trace plots (Cornell 2002, Section
using both the SV and mixture approaches. Table 2
5.9) provide for graphically assessing the effects of
summarizes the results of fitting linear models for two

Table 2: Important Components Identified in Two Screening Examples Using Linear Models

Example Goh & Roy (1989), SV = SiO2 Chan & Kavanagh (1992), SV = Water
Response
Variable Flow Tension Opacity Compression Acidity RMFH GSTT lnVisc lnCLPT
Mixture TiO2, SiO2 TiO2 Na2O, SiO2 SiO2, AEO, AEO, NaCl, No
Approach Na2O Li2O, AEOS, AEOS, NaLAS, components
(CSLM Model) Na2O NaLAS, CDEA, DEALAS, significant
DEALAS, water water
TEALAS,
Water
Slack Variable TiO2, TiO2, TiO2, Na2O, TiO2, Na2O, AEO, AEOS, NaCl, No
Approach Na2O Al2O3, Al2O3 Na2SiF6, CaO, Li2O, AEOS, CDEA NaLAS, components
(Reduced Na2O, Al2O3, ZnO, Al2O3, NaLAS, DEALAS, significant
Linear Model Li2O, CaF2, PbO Na2SiF6 DEALAS, CDEA
by Stepwise Na2SiF6 TEALAS,
Regression)
Important TiO2, TiO2 TiO2 Na2O Na2O, AEO, AEO, N/A N/A
Components Na2O Li2O, AEOS, AEOS,
Identified by Al2O3, NaLAS, NaLAS,
the Original Na2SiF6 DEALAS, DEALAS,
Authors(a) TEALAS, CDEA
CDEA
(a) In some cases authors only identified the component(s) with the largest effects rather than components with significant (i.e.,
non-negligible) effects.

1714
Section on Physical and Engineering Sciences

Table 3: Summary Statistics of Best-Fitting Reduced Quadratic Models for Selected Literature Examples

Cornell and
Example Gorman Fonner et al. (1970) Takayama and Nagai (1991) Guillou and Floros (1993)
(2003) SV = Dicalphos SV = Water SV = Brine Base
SV = x4
Response
Dissolution Y1 tL ln(Rp) TIS Tmin ln(Ymax)
Variable
Mixture Mixture Mixture Mixture
Statistic Mixture SV Mixture SV Mixture SV
and SV and SV and SV and SV
R2 0.978 0.979 0.979 0.886 0.816 0.932 0.918 0.734 0.936 0.932
R2A 0.960 0.977 0.977 0.821 0.746 0.892 0.871 0.586 0.888 0.880
R2P 0.910 0.973 0.973 0.656 0.674 0.720 0.616 0.293 0.761 0.745
RMSE 0.465 0.550 0.541 0.128 0.152 1.858 0.463 10.803 1.758 1.858
Starch2,
Quadratic
Stearic2, stearic dLim× CaCl2× K-sorbate2, K-sorbate2,
Terms in dLim Ethanol2, Ethanol2,
x2 , x3 , x1x3 dicalphos acid2,
2 2
Water, K-sorbate, brine2, NaCl2,
Best ×Eth dLim2 dLim2
×starch starch× Ethanol2 K-sorbate2 CaCl2 CaCl22
Model
stearic

screening examples. Table 3 summarizes the results of compression response. The CSLM model identified Na2O
fitting quadratic models when there were sufficient data to as the only other important component, while the SV
do so for selected quadratic examples. The models in approach identified several other components as
Table 3 contain all linear terms appropriate for mixture or important.
SV models. The last row of Table 3 summarizes the (i)
quadratic terms in reduced quadratic SV models, and (ii) For the Chan and Kavanagh (1992) example, the SV
the selected quadratic terms in PQM models with the (water) has a negative effect on each of four response
mixture approach. variables, as shown in the response trace plots in Figure 2.
For the responses RMFH and lnVisc, the SV approach
3.1 Screening Examples doesn’t identify all of the important components that the
mixture approach (using the CSLM model) identifies.
The mixture and SV approaches reached conflicting
conclusions as to which components were significant in These screening examples show that the SV approach can
the screening examples, as illustrated in Table 2. For make incorrect conclusions such as not recognizing the
some of the responses in the Goh and Roy (1989) SV as an important component and including components
example, the CSLM model identified the SV to be an in the SV linear model that have negligible effects.
important component. The response trace plot in Figure 1
also shows that the SV (SiO2) has a negative effect on the 3.2 Quadratic Examples

There are also several discrepancies between the best


reduced quadratic models that were fit to the quadratic
examples from the literature. For the examples in Table 3,
the mixture approach identified quadratic terms involving
the SV to have significant effects on three responses from
different examples. The SV and mixture experiment
approaches resulted in equivalent models for four of the
responses.

For the tL response in the Takayama and Nagai (1991)


example, the PQM model contains a crossproduct term
involving the SV (water). The SV approach only
identified one significant quadratic term and it didn’t
match the term identified by the PQM approach. The
PQM model was able to achieve a better fit than the SV
Figure 1: Response Trace Plot of Compression Response model, as evidenced by the R 2 , R A2 , R P2 , and RMSE
from Goh and Roy (1989) Example
values shown in Table 3.

1715
Section on Physical and Engineering Sciences

lnCLPT
In the Cornell and Gorman (2003) example, the SV (x4)
1.25 C omponent
does not have a quadratic effect on the Dissolution
nNaC l
nA EO
response, so the PQM and reduced quadratic SV models
1.00 nA EOS
nNaLA S
identified the same quadratic terms and have the same fit.
nDEALAS
nTEA LA S
0.75
Fitted lnCLPT

nC DEA
nW ater
4. Conclusions
0.50

The SV approach makes the assumption that the SV either


0.25
has no effect on the response (e.g., an inactive “filler”
0.00
component) or that it only has a linear additive effect on
the response (e.g., a diluent component). In the literature
-0.050 -0.025 0.000 0.025 0.050 0.075
SV examples analyzed (15 examples with 55 responses,
deviation from reference blend in proportion only some of which are discussed in this paper), we were
unable to find an example of the SV having no effect or
GSTT
an additive linear effect (see Table 1). Hence, these
120 C omponent
nNaC l assumptions appear to be rarely satisfied in practice. The
110
nA EO
nA EOS mixture experiment approach does not make the same
nNaLA S
nDEALAS questionable assumptions, and is applicable whether or
100 nTEA LA S
nC DEA not a particular component has no effect or a linear
Fitted GSTT

90
nW ater
additive effect.
80
The full linear and full quadratic mixture experiment
70
models are equivalent to the SV full linear and full
quadratic models, respectively. When using the full
60
models, the fits of the SV and mixture models (linear or
-0.050 -0.025 0.000 0.025 0.050 0.075
deviation from reference blend in proportion quadratic of each) will be the same but the coefficients
and their levels of significance will not. The advantage to
lnVISC using the mixture approach comes when using (i) CSLM
7 C o mpo nent
nNaC l
models (or Scheffé linear mixture models and response
6 nA EO
nA EO S
trace plots) to assess component linear blending effects
5
nNaLA S
nDEA LA S and (ii) PQM models to understand quadratic blending
effects of components. The SV approach can lead to
nTEA LA S
Fitted lnVISC

nC DEA
4 nW ater
incorrect conclusions about the significance of component
3
linear and quadratic blending effects. Also, the SV models
2 are not capable of including quadratic blending terms
1
involving the SV. This limits the class of quadratic
models possible with the SV approach, which may result
0
-0.050 -0.025 0.000 0.025 0.050 0.075 in an inaccurate representation of the SV (and other
deviation from reference blend in proportion
components) effects.
RMFH
14
C o m p o n en t
The SV and mixture experiment modeling approaches
n N aC l
nA EO often identify different components with significant
13 nA EO S
n N aLA S
n D E A LA S
effects, both with linear models and quadratic models.
12
n TE A LA S
However, the SV and mixture experiment modeling
Fitted RMFH

nC DEA

approaches can also result in equivalent models. This


n W ater

11
should not be taken as evidence of the superiority of the
10 SV approach. It is always possible to obtain a quadratic
mixture model equivalent to a given quadratic SV model
9
by using the PQM modeling methods of the mixture
-0.050 -0.025 0.000 0.025 0.050 0.075
d e v ia t io n fr o m r e fe r e nc e b le nd in p r o po r t io n experiment approach. However, it not always possible to
obtain a given mixture model by using the SV approach.
Figure 2: Response Trace Plots of Each Response This is because the SV approach leaves out the linear SV
Variable in Chan and Kavanagh (1992) Example component and all of the quadratic terms involving the
SV. The mixture approach is more efficient and thorough

1716
Section on Physical and Engineering Sciences

for obtaining an accurate model relating a response Goh, T.N. and Roy, S.K. (1989). “Application of
variable to the component proportions. Taguchi's Orthogonal Array on a Material Screening
Experiment,” Quality Assurance, 15, 10-13.
It is important that researchers be aware of the advantages Guillou, A.A. and Floros, J.D. (1993). “Multiresponse
that mixture models present for the analysis of data from Optimization Minimizes Salt in Natural Cucumber
mixture experiments. The SV approach ignores the effect Fermentation and Storage,” Journal of Food Science,
of the SV on the response, and has fewer potential 58, 1381-1389.
reduced models compared to the class of PQM models. Khuri, A.I. (2005). “Slack-variable Models Versus
The mixture models were specifically designed for Scheffé’s Mixture Models,” Journal of Applied
mixture experiments and provide for obtaining the correct Statistics, 32, 887-908.
analysis and conclusions about linear and quadratic Levison, K.K., Takayama, K., Isowa, K., Okabe, K., and
blending effects of the mixture components. When Nagai, T. (1994). “Formulation Optimization of
analyzing mixture experiments, it is recommended to use Indomethacin Gels Containing a Combination of
the mixture approach rather than the SV approach. Three Kinds of Cyclic Monoterpenes as Percutaneous
Penetration Enhancers,” Journal of Pharmaceutical
Acknowledgements Sciences, 83, 1367-1372.
Piepel, G.F. (2006). “A Note Comparing Component-
Ms. Landmesser’s work was performed as part of a 2007 Slope, Scheffé, and Cox Parameterizations of the
summer internship at the Pacific Northwest National Linear Mixture Experiment Model,” Journal of
Laboratory (PNNL). It was supported by the Department Applied Statistics, 33, 397-403.
of Energy, Office of Science, Scientific Undergraduate Piepel, G.F. (2007). “A Component Slope Linear Model
Laboratory Internship Program. We thank Scott Cooley of for Mixture Experiments,” Quality Technology and
the Statistical Sciences group at PNNL for reviewing a Quantitative Management, 4, 331-343.
draft of this paper. Piepel, G.F. and Cornell, J.A. (1994). “Mixture
Experiment Approaches: Examples, Discussion, and
References Recommendations,” Journal of Quality Technology,
26, 177-196.
Abdullah, A., Resurreccion, A.V.A., and Beuchat, L.R. Piepel, G.F. and T. Redgate (1997). “Mixture Experiment
(1993). “Formulation and Evaluation of a Peanut Techniques for Reducing the Number of Components
Milk Based Whipped Topping Using Response Applied to Modeling Waste Glass Sodium Release,”
Surface Methodology,” Lebensmittel-Wissenschaft Journal of the American Ceramic Society, 80, 3038-
und-Technologie (Food Science and Technology), 26, 3044.
162-166. Piepel, G.F., Szychowski, J.M., and Loeppky, J.L. (2002).
MacDonald, I.A. and Bly, D.A. (1966). “Determination of “Augmenting Scheffé Linear Mixture Models with
Optimal Levels of Several Emulsifiers in Cake Mix Squared and/or Crossproduct Terms,” Journal of
Shortenings.” Cereal Chemistry, 43, 571-584. Quality Technology, 34, 297-314.
Cain, M. and Price, M.L.R. (1986). “Optimal Mixture Scheffé, H. (1958), “Experiments with Mixtures,” Journal
Choice,” Applied Statistics, 35, 1-7. of the Royal Statistical Society, B, 20, 344-360.
Chan, K.Y. and Kavanagh, P.E. (1992). “Application of Setz, S., Semling, M. and Mulhaupt, R. (1997). “Fuzzy
Plackett-Burman Design and Linear Programming to Set Approach for Fitting a Continuous Response
Light-Duty Liquid Detergent Formulation,” Journal Surface in Adhesion Formulation,” Journal of
of the American Oil Chemists Society, 69, 690-694. Chemometrics, 11, 403-418.
Cornell, J.A. (2000). “Fitting a Slack-Variable Model to Smith, W.F. (2005). Experimental Design for
Mixture Data: Some Questions Raised,” Journal of Formulation. ASA-SIAM Series on Statistics and
Quality Technology, 32, 133-147. Applied Probability, SIAM, Philadelphia, PA and
Cornell, J.A. (2002). Experiments with Mixtures: Designs, ASA, Alexandria, VA.
Models and the Analysis of Mixture Data, Third Takayama, K. and Nagai, T. (1991). “Simultaneous
Edition, John Wiley and Sons, New York. Optimization for Several Characteristics Concerning
Cornell, J.A. and Gorman, J.W. (2003). “Two New Percutaneous Absorption and Kin Damage of
Mixture Models: Living With Collinearity but Ketoprofen Hydrogels Containing d-Limonene,”
Removing Its Influence,” Journal of Quality International Journal of Pharmaceutics, 74, 115-126.
Technology, 35, 78-88.
Fonner, D.E., Buck, J.R., and Banker, G.S. (1970).
“Mathematical Optimization Techniques in Drug
Product Design and Process Analysis,” Journal of
Pharmaceutical Sciences, 59, 1587-1596.

1717

S-ar putea să vă placă și