1 Response Surface A Response surface model is a special type of multiple regression model with: Explanatory variables Interaction variables Squared variables.

1

Response Surface A Response surface model

is a special type of multiple regression model with: Explanatory variables Interaction variables Squared variables

2

Response Surface A response surface model is

often used to approximate a complicated relationship between a response variable and several explanatory variables.

3

Response Surface In order to get reliable data

for a response surface model, a designed experiment is often used to collect the data on the explanatory variables and the response.

4

Designed Experiment In a designed experiment,

the experimenter chooses values of the explanatory variables to investigate and measures the response for the chosen combinations of explanatory variables.

5

Tennis Ball Experiment In the manufacture of tennis

balls certain additives are thought to affect the bounciness of the tennis ball.

Response: A measure of bounce.

6

Tennis Ball Experiment Explanatory variables

Amount of silica Amount of sulfur Amount of silane

7

Tennis Ball Experiment Each explanatory variable

has three levels Silica: 0.7, 1.2, 1.7 Sulfur: 1.8, 2.3, 2.8 Silane: 40, 50, 60

8

Tennis Ball Experiment A total of 15 combinations of

silica, sulfur and silane are examined and the bounce response is measured for each combination.

The target bounce response is 450.

9

Silica Sulfur Silane Bounce

0.7 1.8 50 5700.7 2.8 50 2851.7 1.8 50 2601.7 2.8 50 4331.2 1.8 40 422

1.2 2.3 50 396

10

Response Surface Model

2333

2222

2111

322331132112

3322110

xxx

xxxxxxxxxY

11

JMP – Fit Model Put Bounce in for the Y

response. Highlight silica, sulfur and

silane in Select Columns. Macros – Response Surface

12

13

14

Summary The model is useful.

F=2488.146, P-value < 0.0001

R2=0.999777, virtually all of the variation in Bounce is explained by the model.

15

16

Statistical Significance The interaction between

Silica and Silane is not statistically significant.

The squared term for Silane is not statistically significant.

17

Reduced Model Remove the interaction

term: Silica*Silane. Remove the squared term:

Silane*Silane.

18

19

Summary The model is useful.

F=4372.207, P-value < 0.0001 R2=0.999771, virtually all of

the variation in Bounce is explained by the model. Only slightly lower than R2 for the full model.

20

Statistical Significance All variables in the model

are statistically significant. This is the best response

surface model.

21

Prediction What combinations of silica,

sulfur and silane will give you the target bounce of 450, on average?

22

23

24

Prediction There are several

combinations that will give a predicted bounce of 450. Silica = 1.0, Sulfur = 1.948, Silane = 50

Silica = 0.8, Sulfur = 2.251, Silane = 40