Scribd Logo
Upload

Welcome to Scribd!

  • 26 views

    Model Answers Part B 2008

    Uploaded by

    Pi
    This document provides model answers for part B of a final exam in the STATS 330/762 course. 1. The best model for a dataset with factors and continuous variables is one with interactions between all factors and continuous variables. However, there may not be enough data points for such a complex model. Stepwise regression and information criteria like AIC can help select a better model. 2. Logistic regression models the log-odds of an event using explanatory variables. Coefficients represent the change in log-odds. Interactions have no simple interpretation and require examining a table of log-odds. 3. Odds ratios can test independence between factors in a contingency table. A confidence interval around the

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Model Answers Part B 2008

    Uploaded by

    Pi
    26 views2 pages
    This document provides model answers for part B of a final exam in the STATS 330/762 course. 1. The best model for a dataset with factors and continuous variables is one with interactions between all factors and continuous variables. However, there may not be enough data points for such a complex model. Stepwise regression and information criteria like AIC can help select a better model. 2. Logistic regression models the log-odds of an event using explanatory variables. Coefficients represent the change in log-odds. Interactions have no simple interpretation and require examining a table of log-odds. 3. Odds ratios can test independence between factors in a contingency table. A confidence interval around the

    Original Description:

    Model Answers Part b 2008

    Original Title

    Model Answers Part b 2008

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document provides model answers for part B of a final exam in the STATS 330/762 course. 1. The best model for a dataset with factors and continuous variables is one with interactions between all factors and continuous variables. However, there may not be enough data points for such a complex model. Stepwise regression and information criteria like AIC can help select a better model. 2. Logistic regression models the log-odds of an event using explanatory variables. Coefficients represent the change in log-odds. Interactions have no simple interpretation and require examining a table of log-odds. 3. Odds ratios can test independence between factors in a contingency table. A confidence interval around the

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    26 views2 pages

    Model Answers Part B 2008

    Uploaded by

    Pi
    This document provides model answers for part B of a final exam in the STATS 330/762 course. 1. The best model for a dataset with factors and continuous variables is one with interactions between all factors and continuous variables. However, there may not be enough data points for such a complex model. Stepwise regression and information criteria like AIC can help select a better model. 2. Logistic regression models the log-odds of an event using explanatory variables. Coefficients represent the change in log-odds. Interactions have no simple interpretation and require examining a table of log-odds. 3. Odds ratios can test independence between factors in a contingency table. A confidence interval around the

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 2

    Department of Statistics

    COURSE STATS 330/762

    Final exam 2008: Model answers for part B

    1. (a) A model with interactions between all the factors and each continuous
    variable e.g. for factors A, B and continuous variables X,W we would fit the
    model A*B*X + A*B*W. There might not be enough data to fit such a model
    (requires enough points for each possible combination of the factor levels)

    (b) Stepwise regression, use of anova function, AIC.

    (c) All criteria choose model 6. Models 5 and 7 are not much worse on CV
    grounds

    2. (a) The logistic model with response r/n, where r has a binomial distribution
    with parameters n and , and logit () is a linear combination of the
    explanatory variables.

    (b) For continuous covariates, b is the change in log-odds corresponding to a


    unit increase in the covariates, assuming all other covariates are held
    constant. For factors, b is the change in the log-odds when the level is
    changed from the baseline (assuming main effects only). When there are
    interactions, there is no simple explanation, you must make a table of log-
    odds to assess the effect of changing the factor levels.

    (c) Note: the regression table was omitted unfortunately I no longer have a
    copy.

    (d) A confidence interval for the log-odds is -2.272071 +/- 1.96* 0.5421498 or
    (-3.3347, -1.209457). Applying the function exp(x)/(1+exp(x)) to both ends of
    this interval we get (0.0344, 0.2298).

    3. (a) Odds ratio is (assuming row facor in the 2x2 table is A and the column
    factor is B)
    (odds of baseline of A at baseline level of ) / (odds of baseline of A at other level of B)

    If the table is

    1
    B
    0 1
    0 1 2
    A
    1 3 4

    then the OR is (1x 4)/( 2x 3).

    The two factors are independent if the odds ratio is 1. To find a CI


    for the odds ratio we fit a Poisson regression count ~ A *B (the maximal
    model) and use the interaction line in the summary table to get the estimate
    and standard error of the log OR. An interval for the log OR is estimate +/-
    1.96* std error. Exponentiating both ends gives a CI for the OR.

    (b) The odds ratio s are the same when the 3-factor interaction is zero. To test
    if the 3-factor interaction is zero, fit the maximal model count~A*B*C and use
    the p-value in the 3-way interaction line in the summary table.

    (c) The Defendants race/Death penalty interaction is 0.3873 with a std error
    of 0.2785. The log OR confidence interval is 0.3873 +/- 1.96*0.2785 or
    ( -0.1586 , 0.9332). Exponentiating gives (0.8534, 2.5425). Since this contains
    1 the data are consistent with the OR being 1 (i.e. conditional independence
    of defendants race and death penalty, given victims race.)

    You might also like