Structured Assignment One Solutions
Objective:
The purpose of this assignment is to help you understand the basics of experimental design,
linear model formulation, and data table creation.
Instructions:
Complete all items below. There are two process checkpoints shown in bold. At both points, ask
an instructor to review your work so you can get credit for your progress.
1. Choose a theme for your experimental design. For example, you can choose making
lemonade or performance of a brine evaporator.
2. Define an outcome variable, Y. For lemonade you might define Y to be the “tastiness” of
the lemonade on a 0 to 10 scale. For a brine evaporator, you might choose volume of
water evaporated per unit time.
3. Choose two design factors that you want to test. For example, you can choose amount
of sugar and amount of lemon juice in lemonade. For a brine evaporator, you can
choose spacing of evaporative fins and thickness of fabric covering the fins.
4. Identify three levels of each factor. For example, you can choose low, medium, and high.
Define these clearly (e.g., low = 10 g sugar, medium = 20 g sugar, etc.)
5. Create a two-factor experimental design table based on your inputs in Steps 1 to 3
above. Show each experiment as its own row. Each column should be a factor. The
entries in the table indicate the level of a given factor. Fill in the table below for a full
factorial experimental design (show all permutations of experimental factors). Ask an
instructor to check your work and give you credit.
Experiment # Factor 1 Factor 2
1
2
3
4
5
6
7
8
9
6. Suppose separate binary (0,1) indicator variables are to be coded for each level of each
factor with 0 indicating the factor is not set to the given level, and 1 indicating that the
factor is set to the given level. Think about an equation (or mathematical model) in
which Y is a function of the different experimental design factors:
𝑌𝑌 = 𝛽𝛽0 + 𝛽𝛽1,ℎ𝑖𝑖𝑖𝑖ℎ 𝐼𝐼1,ℎ𝑖𝑖𝑖𝑖ℎ + 𝛽𝛽1,𝑙𝑙𝑙𝑙𝑙𝑙 𝐼𝐼1,𝑙𝑙𝑙𝑙𝑙𝑙 … .
where 𝛽𝛽 values are parameters estimated from the data (units match units of Y), k is the
number of explanatory variables in the model, and the I values are the 0,1 indicator variables
(dimensionless). The simplest way to start is to consider just the impacts of different levels of
the factors individually. These are called the “main effects”. Choose “medium” to be the
“baseline” condition. The baseline condition is the condition for which there is no indicator
variable. Extend the equation above to show main effects for all your experimental factors. This
means you should include indicator variables for when each factor is high and when each factor
is low. Ask an instructor to check your work and give you credit.
7. What does the intercept, 𝛽𝛽0 represent?
8. What does the parameter𝛽𝛽1,ℎ𝑖𝑖𝑖𝑖ℎ represent?
9. Think about the I parameters, or indicator variables, in the linear equation above. These
are “data”, that is, they are not estimated but are specified as either 0 or 1 depending
on the conditions under which the experiment was run. Each row in the table you
developed for Question 5 will have a unique combination of these indicator variables.
Make a table showing the values for each of the 4 different indicator values for each of
the experiments listed in the table you developed for Question 5. The rows will be the
same as the rows from Step 4 (i.e., each row indicates a particular experiment).
However, this table should have each indicator variable as a column heading. The
entries in the table will be 1 when the indicator variable is set to the indicated level and
0 otherwise. For example, the first experiment should have low levels of both factors.
The first row of the table you develop for this question should have column entries that
show 𝐼𝐼1,𝑙𝑙𝑙𝑙𝑙𝑙 =1, 𝐼𝐼1,ℎ𝑖𝑖𝑖𝑖ℎ =0, 𝐼𝐼2,𝑙𝑙𝑙𝑙𝑙𝑙 =1, 𝐼𝐼2,𝑙𝑙𝑙𝑙𝑙𝑙 =0.
Experiment # 𝐼𝐼1,𝑙𝑙𝑙𝑙𝑙𝑙 𝐼𝐼1,ℎ𝑖𝑖𝑖𝑖ℎ 𝐼𝐼2,𝑙𝑙𝑙𝑙𝑙𝑙 𝐼𝐼2,ℎ𝑖𝑖𝑖𝑖ℎ
1 1 0 1 0
2
3
4
5
6
7
8
9
10. In some cases the effect of one level may be modified by (or interact with) the level of
another factor. For example, we might dislike increasing lemon from baseline if sugar is
held constant, but we might like the additional lemon when sugar is increased. We
denote interactions by taking the product of the indicator variables and including a new
𝛽𝛽 for the product. For example, we could include an interaction for both factors being
high by adding a term to our equation showing an effect when both factors are high, as
follows:
𝑌𝑌 = 𝛽𝛽0 + 𝛽𝛽1,ℎ𝑖𝑖𝑖𝑖ℎ 𝐼𝐼1,ℎ𝑖𝑖𝑖𝑖ℎ + 𝛽𝛽1,𝑙𝑙𝑙𝑙𝑙𝑙 𝐼𝐼1,𝑙𝑙𝑙𝑙𝑙𝑙 + 𝛽𝛽2,ℎ𝑖𝑖𝑖𝑖ℎ 𝐼𝐼2,ℎ𝑖𝑖𝑖𝑖ℎ + 𝛽𝛽2,𝑙𝑙𝑙𝑙𝑙𝑙 𝐼𝐼2,𝑙𝑙𝑙𝑙𝑙𝑙 + 𝛽𝛽1ℎ𝑥𝑥2ℎ 𝐼𝐼1,ℎ𝑖𝑖𝑖𝑖ℎ
∗ 𝐼𝐼2,ℎ𝑖𝑖𝑖𝑖ℎ
The effect parameter 𝛽𝛽1ℎ𝑥𝑥2ℎ can be thought of as a “bonus” for when both factors are at their
highest levels (of course the bonus can be positive or negative). An interaction can occur
between any two indicator variables that can both be 1 at the same time. For example,
𝛽𝛽1,ℎ𝑖𝑖𝑖𝑖ℎ 𝐼𝐼1,ℎ𝑖𝑖𝑖𝑖ℎ cannot interact with 𝛽𝛽1,𝑙𝑙𝑙𝑙𝑙𝑙 𝐼𝐼1,𝑙𝑙𝑙𝑙𝑙𝑙 because they are never both 1 (an
experiment cannot simultaneously have two levels of the same factor). While a variable
cannot interact with an indicator for the same factor, it can interact with all indicators
for other factors. Extend the equation above to include all possible interactions.
11. The method above estimates main effects and interactions. Another approach would be
to simply estimate a mean for each unique experiment. How many unique means would
there be? How many parameters are in the linear model with the main effects and
interactions?
