Scribd Logo
Upload

Welcome to Scribd!

  • 33 views

    Department of Statistics: Course Stats 330

    Uploaded by

    PETER
    This document contains model answers for a 2004 statistics final exam. It includes: 1) The multiple choice answers to questions 1 through 20. 2) Summaries of the answers to three questions in Section B about model fitting, risk factor analysis, and Simpson's paradox. 3) The summaries identify key patterns in the data, appropriate statistical models, important factors and their effects, and how to present results visually.

    Copyright:

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

    Department of Statistics: Course Stats 330

    Uploaded by

    PETER
    33 views4 pages
    This document contains model answers for a 2004 statistics final exam. It includes: 1) The multiple choice answers to questions 1 through 20. 2) Summaries of the answers to three questions in Section B about model fitting, risk factor analysis, and Simpson's paradox. 3) The summaries identify key patterns in the data, appropriate statistical models, important factors and their effects, and how to present results visually.

    Original Description:

    04exam Ans

    Original Title

    04exam Ans

    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 contains model answers for a 2004 statistics final exam. It includes: 1) The multiple choice answers to questions 1 through 20. 2) Summaries of the answers to three questions in Section B about model fitting, risk factor analysis, and Simpson's paradox. 3) The summaries identify key patterns in the data, appropriate statistical models, important factors and their effects, and how to present results visually.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    33 views4 pages

    Department of Statistics: Course Stats 330

    Uploaded by

    PETER
    This document contains model answers for a 2004 statistics final exam. It includes: 1) The multiple choice answers to questions 1 through 20. 2) Summaries of the answers to three questions in Section B about model fitting, risk factor analysis, and Simpson's paradox. 3) The summaries identify key patterns in the data, appropriate statistical models, important factors and their effects, and how to present results visually.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 4

    Department of Statistics

    COURSE STATS 330

    Model answers for Final Exam, 2004

    Multiple Choice:

    Question Answer Question Answer


    1 3 11 1
    2 1 12 1
    3 5 13 5
    4 5 14 3
    5 3 15 2
    6 4 16 5
    7 3 17 4
    8 1 18 2
    9 2 19 4
    10 4 20 2

    Section B

    Q1
    a) The indicated model fits six parallel lines, one to each panel of the coplot. This seems more
    or less consistent with the patterns in the coplot, particularly for the panels where the data
    extend across the width of the panel.

    b) The output suggests that the there is no interaction between HeadWt and the other factors.
    This confirms that the parallel line model in (a) is reasonable. In fact an even simpler
    model, ignoring Date, also seems reasonable. We would want do a test to confirm this.

    c) The model here is fitting 6 arbitrary (i.e. not necessarily parallel) lines, one to each panel.
    The large dfbetas all correspond to terms determining the slope and intercept of the line
    corresponding to Line 52, Date 20, the middle plot in the top row. The influential points
    are the ones whose removal would have a big change on the line. These will probably be
    the points at the extreme ends of the weight range in that panel. Take your pick!

    d) In this new model the lines are now constrained to have equal slopes. All 60 points now
    contribute to the estimation of this slope. Removing one point out of 60 will have a smaller
    effect than removing 1 point out of 10.

    e) There is a strong suggestion that date has no effect on the response, as all terms involving
    data have insignificant p-values. Line 52 cabbages have significantly higher contents. This

    1
    effect does depend on HeadWt, as this is highly significant in the summary. The heavier the
    head, the lower the ascorbic acid content, for a fixed line.

    Q2

    a) Initially fit a saturated model of the form

    cbind(num,tot-num) ~ sex*serum*age

    Use anova, stepwise to select a suitable submodel. Can use the residual deviance to check fit of
    a submodel, and diagnostic plots to see which factor-level combinations are poorly fit by the
    chosen model.

    b) Inspection of the diagnostic plots reveals that points 5 and 13 are poorly fitted by the model.
    Inspection of the data reveals that the rates for males are at least twice the rates for females,
    except for 5 and 13 (age group 50-62, Serum A) This suggests that the male and female data
    may have been transposed for these two groups.

    c) To discuss the rates and the effect of the factors, a graph is helpful:

    This could be drawn by hand, eg one of the form

    Rate

    A B C D

    with one line for each age-sex combination, using the predicted data supplied. We can then see
    that
    age is the most important factor (has the biggest coefficient in the summary)
    Sex is the second most important factor, followed by serum

    2
    As age increases, probability of CHD goes up.
    Males have higher risk than females
    Serum A and serum B about equal, but C and D have higher probabilities, with D higher
    than C

    3
    Q3

    a) and b) taken from lecture on Simpsons paradox.

    c) In the marginal table, collapsing over Departments, there is a strong association between
    Gender and Admission. The p-value is tiny corresponding to 93.45 on 1 df.

    d) When Department is taken into account, the gender:Admit and 3-factor interactions are
    insignificant, suggesting that gender and admission are conditionally independent, given
    department. Thus the conclusion from the marginal table is different from that based on the
    conditional tables. This is because the condition for collapsing over Depts is not met (the
    Dept-gender and dept-admit interactions are highly significant).

    You might also like