Programming Assignment 3

CSC 323 Data Analysis and Statistical Software

Due: 6/4/98

Use simple linear regression methods to conduct a thorough analysis of the following dataset.

• Fault Dataset:

  Each observation consists of the following variables in the order presented:

  • Fault Detection Effectiveness
  • Decision Coverage
  • Block Coverage
  • All-Uses Coverage
  • Utility Name (Unix utilities Cal & Col)

  Consider Fault Detection Effectiveness to be the dependent variable. Decision, Block and All-Uses Coverage are explanatory variables.

An example of the format required for your analysis is provided. However, your analysis will contain more. It will have a section which identifies the best explanatory variable and your Discussion section will discuss the 95% confidence intervals for your parameter estimates.

Your analysis should address the following points:

• Identify the best explanatory variable, that is, the explanatory variable that does the best job of explaining variability in the dependent variable.
• For the best explanatory variable:
  • Identify the regression model including symbols for all of its parameters.
  • Conduct a residual analysis. You must state the hypotheses and p-value for the normality test.
  • Give the estimates of the model parameters, state the regression equation and interpret the coefficients of the equation.
  • Construct, and interpret, 95% confidence intervals for the slope and intercept parameters.
• Summarize the results of your analysis.

Your program should accomplish the following:

• Produce a scatterplot of the dependent variable vs. each of the explanatory variables.
• Execute PROC CORR for the dependent variable and each of the explanatory variables.
• For the best explanatory variable:
  • Generate estimates of your slope and intercept using PROC REG.
  • Produce a residual plot.
  • Execute PROC UNIVARIATE with the NORMAL and PLOT options for the residuals.
  • Optionally produce a normal plot of the residuals using the RANK and PLOT procedures.
    Note:You may want to do this if the normality test is not definitive.