Case LR3: Preference Testing - Introduction to Predicting ...



Study Case: Preference Testing

Following exercises are from the study by Ries and Smith. (Chemical Engineering Progress, 59, 1963: 39-43).

An improved detergent, say X, has been formulated. It is considered a much better product than its old form, say M. The question is to test whether or not users prefer X over M in a blind test. For a random sample of 1008 consumers the following questions are asked.

1. Which of the products did you prefer?

2. What laundry detergent do you generally use?

3. What kind of water is used in your locality?

4. What is the temperature of the laundry water used?

TABLE 1

Cross-Classification of Sample of 1008 Consumers According to (1) the Softness of the Laundry Water Used, (2) the Previous Use of Detergent Brand M, (3) the Temperature of the Laundry Water Used, (4) the Preference for Detergent Brand X over Brand M in a Consumer Blind Trial

Previous User of M Previous Non-User of M

Water Brand High Low High Low

Softness Preference Temperature Temperature Temperature Temperature

Soft X 19 57 29 63

M 29 49 27 53

Medium X 23 47 33 66

M 47 55 23 50

Hard X 24 37 42 68

M 43 52 30 42

Weighted Least Squares Estimation Using SPSS

W = 1, 2, and 3 for soft to hard

Use = 1 if previous use of M, 0 otherwise

TH = 1 if temperature is high and 0 if low.

1. Full model

2. Reduced Model

Appendix: Logistic Regression Using Weighted Least Squares

1. Variance of Disturbances

Let:

R = Number of Successes in n independent trials, with probability of

success at each trial p.

R follows a binomial distribution (n, π).

E(R) = n π

σ(R) = [pic]

When n is large so that n π >5 and n(l-π) >5, R is approximately normal.

Let:

[pic]

Then:

(1) [pic] follows approx. N[pic]

(2) logit[pic]follows approx. N[pic]

For logistic regression using grouped data, estimate:

[pic]

2. Weighted Least Squares

Let the population regression model be defined as follows:

[pic] where [pic] follows random [pic]

Standard case

[pic]

B. Variances are different but known

The estimates of the regression coefficients are determined by the weighted least squares.

Minimize [pic] where [pic]

Define a diagonal matrix of weights, W as follows:

[pic]

Then [pic][pic]

.

[pic]

Note:

It is important to note that b follows normal distribution.

Here are key ANOVA results:

[pic]

[pic] where [pic] the weighted average of Y.

[pic]

For testing significance of the sample regression, we use the result:

Under [pic]

SSR follows a [pic] distribution with DF = K.

C. Variances are proportional to known values

[pic]

b is determined by the formula given in B. However,

[pic]

where [pic]

Note:

The weighted least squares routine in standard statistics packages assume this variance error model for estimation.

-----------------------

Use [pic] distribution DF=3 for computing the p-value.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download