Example of Analyze Binary Response for Response Surface Design

A clean room engineer analyzes a response surface design to determine how seal time, temperature, and pressure affect the seal quality of the sealed trays. The response is binary—whether the seal is intact or not—in a sample of 800 tray seals.

The engineer collects data and analyzes the design to determine which factors impact seal strength.

  1. Open the sample data, TraySeal.MTW.
  2. Choose Stat > DOE > Response Surface > Analyze Binary Response.
  3. In Event name, enter Event.
  4. In Number of events, enter Sealed.
  5. In Number of trials, enter Samples.
  6. Click Terms.
  7. Under Include the following terms, choose Full quadratic.
  8. Click OK.
  9. Click Graphs.
  10. Under Residual Plots, select Four in one.
  11. Click OK in each dialog box.

Interpret the results

In the Analysis of Variance table, the p-values for Temperature, Pressure and Temperature*Temperature are significant. The engineer can consider reducing the model to remove the terms that are not significant. For more information, go to Model reduction.

The Deviance R2 value shows that the model explains 97.47% of the total deviance in the response, which indicates that the model fits the data well.

The Pareto plot of the effects allow you to visually identify the important effects and compare the relative magnitude of the various effects. In addition, you can see that the largest effect is Temperature*Temperature (BB) because it extends the farthest.

Method

Link functionLogit
Rows used15

Response Information

VariableValueCountEvent Name
SealedEvent9637Event
  Non-event2363 
SamplesTotal12000 

Coded Coefficients

TermCoefSE CoefVIF
Constant3.0210.384 
Time0.2100.13918.53
Temperature0.6410.15919.53
Pressure0.4200.21170.48
Time*Time-0.07350.04821.01
Temperature*Temperature0.29880.05171.17
Pressure*Pressure-0.00220.027770.24
Time*Temperature-0.00920.05051.14
Time*Pressure0.04170.034218.12
Temperature*Pressure-0.05210.039619.24

Odds Ratios for Continuous Predictors

Unit of
Change
Odds Ratio95% CI
Time1.0*(*, *)
Temperature25.0*(*, *)
Pressure7.5*(*, *)
Odds ratios are not calculated for predictors that are included in interaction terms because
     these ratios depend on values of the other predictors in the interaction terms.

Model Summary

Deviance
R-Sq
Deviance
R-Sq(adj)
AICAICcBIC
97.47%96.50%140.64195.64147.72

Goodness-of-Fit Tests

TestDFChi-SquareP-Value
Deviance523.400.000
Pearson523.880.000
Hosmer-Lemeshow57.470.188

Analysis of Variance

SourceDFAdj DevAdj MeanChi-SquareP-Value
Model9903.478100.386903.480.000
  Time12.3032.3032.300.129
  Temperature116.38816.38816.390.000
  Pressure13.9663.9663.970.046
  Time*Time12.3312.3312.330.127
  Temperature*Temperature134.01234.01234.010.000
  Pressure*Pressure10.0060.0060.010.937
  Time*Temperature10.0330.0330.030.856
  Time*Pressure11.4901.4901.490.222
  Temperature*Pressure11.7311.7311.730.188
Error523.4044.681   
Total14926.882     

Regression Equation in Uncoded Units

P(Event)=exp(Y')/(1 + exp(Y'))
Y'=17.77 + 0.348 Time - 0.1918 Temperature + 0.1146 Pressure - 0.0735 Time*Time
+ 0.000478 Temperature*Temperature - 0.000039 Pressure*Pressure
- 0.00037 Time*Temperature + 0.00556 Time*Pressure - 0.000278 Temperature*Pressure

Fits and Diagnostics for Unusual Observations

ObsObserved
Probability
FitResidStd Resid
10.71130.68561.57224.45R
30.90250.88791.33702.50R
70.96750.95651.59272.17R
80.67370.6884-0.8891-2.44R
100.55500.5660-0.6265-2.07R
110.90250.9281-2.6700-4.20R
120.84130.8633-1.7806-3.54R
150.71130.68921.35923.64R
R  Large residual