Search code examples
rmatlabfactorialanovaminitab

How to calculate percentage change using factorial analysis?

To assess the contribution of various input factors (x, y, z) and their interactions to the response variable (A), I calculated the percentage of variance using factorial analysis in Minitab. Now I want to calculate the percentage change in A.

Say for example A is increasing when x and y are increasing and A is decreasing when z is decreasing. So how much percentage is changing in A? Is there any other software by which I can do this analysis?

Please help me in this regards.

Thanks.

Solution

  • I think that a simple regression would provide you what you are looking for. Since you are asking how to perform this computation for both R and Matlab, I'll provide you a Matlab solution since I can use it much better.

    Before proceeding with a numerical example, let's review a little bit of theory:

    A fitted linear regression model can be used to identify the relationship between a single predictor variable xj and the response variable y when all the other predictor variables in the model are "held fixed". Specifically, the interpretation of βj is the expected change in y for a one-unit change in xj when the other covariates are held fixed—that is, the expected value of the partial derivative of y with respect to xj.

    Basically, this tells us that a one unit change in xj generates a βj unit change in y. In order to obtain percent changes, the response variable y must be converted to a logarithmic scale (ln(y)).

    Now, let's see how to perform a linear regression using the regress function in Matlab. This is as simple as:

    % Response Variable
A = rand(100,1);

% Predictors
X = randi(10,100,1);
Y = rand(100,1);
Z = randi(3,100,1);

% Beta Coefficients
b = regress(A,[X Y Z]);

    Now, in order to retrieve percent changes instead of unit changes, the above code must be rewritten as follows (basically, a natural log is applied to A and percent changes are calculated multiplying the beta coefficients b by 100):

    % Response Variable
A = rand(100,1);
A = log(A);

% Predictors
X = randi(10,100,1);
Y = rand(100,1);
Z = randi(3,100,1);

% Beta Coefficients
b = regress(A,[X Y Z]);

% Percent Changes
pc = b .* 100;

    Using arbitrary vales, let's suppose that the returned betas are:

    b =

    0.25
   -0.06
    1.33

    This means that: a one unit change in X generates a +25% change in A, a one unit change in Y generates a -6% change in A and a one unit change in Z generates a +133% change in A. The interpretation of percent changes varies following the type of the predictor variables, and you must be careful about this. Given the response variable Y and the predictor variable K:

    • If K is a continuous variable, then Bk = ∂ln(Y) / ∂K, hence a one unit change in K generates a 100 * Bk percent change in Y.
    • If K is the natural log of a continuous variable, then Bk = ∂ln(Y) / ∂ln(K), hence a 100% change in K generates a 100 * Bk percent change in Y.
    • If K is a dummy variable (only two possible values: 1 = true and 0 = false), then the shift of K from 0 to 1 produced a 100 * Bk percent change in Y.