Scribd Logo
Upload

Welcome to Scribd!

  • 4 views

    Chapter 6

    Uploaded by

    yonasante2121

    Copyright:

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

    Chapter 6

    Uploaded by

    yonasante2121
    4 views24 pages

    Original Title

    Chapter 6 - Copy

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    4 views24 pages

    Chapter 6

    Uploaded by

    yonasante2121

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 24

    Chapter 6: generalized linear models (GLMs)

    for count data:


    Poisson regression and Loglinear Models
    By: Ashenafi A.
    ➔ Introduction to Poisson regression

    ➔ Estimation and testing

    ➔ Count Regression for Rate Data

    ➔ Loglinear Models for Two-Way and Three-Way

    Tables,

    ➔ Loglinear Model of Independence

    Overview ➔ Interpretation of Parameters in Independence

    Model,

    ➔ Saturated Model

    ➔ ,
    Poisson regression
    ➢ Many discrete response variables have counts as possible
    outcomes. Example is Y = number of parties attended in the
    past month, for a sample of students,
    ➢ Counts also occur in summarizing categorical variables with
    contingency tables.
    ➢ This section introduces GLMs for count data
    ➢ The simplest GLMs for count data assume a Poisson
    distribution for the random component. Like counts, Poisson
    variates can take any nonnegative integer value.
    Poisson Model
    The Maximum Likelihood estimator
    ➔ We observe data {(xi , yi)|1 ≤ i ≤ n}. The number yi is a realization of the random
    variable Yi . The total log-likelihood is, using independency, given by

    and 𝛍i = exp(βtxi). Write now Log L(β) as shorthand notation for the total likelihood. Then it follows

    The maximum likelihood (ML) estimator is then of course defined as


    Poisson regression
    ❖ For a single explanatory variable x, the Poisson loglinear
    model has form

    A one-unit increase in x has a multiplicative impact of eβ on μ:


    Interpretation of β (continued)
    Example: Number of Deaths Due to AIDs
    The number of deaths due to AIDS per 3-month period from January 1983 to
    June 1986
    Number of Deaths Due to AIDs × Month
    y<-c(0,1,2,3,1,4,9,18,23,31,20,25,37,45)
    x<-c(1,2,3,4,5,6,7,8,9,10,11,12,13,14)
    plot(x,y ,col= "red")
    A Linear Model for AIDs Data
    Let’s try a linear model:
    The estimated parameters from GLM with a Poisson distribution and the log link:
    R-code
    > poisson.model<-glm(y ~ x, family = quasipoisson(link = "log"))
    > summary(poisson.model)
    Call:
    glm(formula = y ~ x, family = quasipoisson(link = "log"))
    Coefficients:
    Estimate Std. Error t value Pr(>|t|)
    (Intercept) 0.33963 0.38946 0.872 0.4
    x 0.25652 0.03417 7.507 7.17e-06 ***
    ---
    Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

    (Dispersion parameter for quasipoisson family taken to be 2.403942)

    Null deviance: 207.272 on 13 degrees of freedom


    Residual deviance: 29.654 on 12 degrees of freedom
    AIC: NA
    Number of Fisher Scoring iterations: 5
    Example
    The number of awards earned by students at one high school.
    Predictors of the number of awards earned include the type of
    program in which the student was enrolled (e.g., vocational,
    general or academic) and the score on their final exam in math.
    Data available in this link

    : https://stats.idre.ucla.edu/wp-content/uploads/2016/02/poisson_sim.sav.
    >p<-poisson_sim
    >hist(p$num_awards,col
    = "yellow",border =
    "blue",xlim = c(0,8),
    ylim = c(0,200),
    breaks = 7,main =
    "Histogram",xlab =
    "Number of award")
    con’t
    > summary(m1 <- glm(num_awards ~ prog + math, family="poisson", data=p))

    Call:
    glm(formula = num_awards ~ prog + math, family = "poisson", data = p)
    Deviance Residuals:
    Min 1Q Median 3Q Max
    -2.2043 -0.8436 -0.5106 0.2558 2.6796
    Coefficients:
    Estimate Std. Error z value Pr(>|z|)
    (Intercept) -5.24712 0.65845 -7.969 1.60e-15 ***
    progAcademic 1.08386 0.35825 3.025 0.00248 **
    progVocational 0.36981 0.44107 0.838 0.40179
    math 0.07015 0.01060 6.619 3.63e-11 ***
    ---
    Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
    (Dispersion parameter for poisson family taken to be 1)
    Null deviance: 287.67 on 199 degrees of freedom
    Residual deviance: 189.45 on 196 degrees of freedom
    AIC: 373.5
    Number of Fisher Scoring iterations: 6
    Interpret
    log(num_awards) = Intercept + b1( progAcademic) +
    b2(progVocational ) + b3math.
    = exp(Intercept) * exp(b1(progAcademic)) *
    exp(b2(progVocational)) * exp(b3math)
    The output above indicates that the award rate for [prog=Academic] is
    2.9560 times the award rate for the reference group, general. Likewise,
    the award rate for [prog=Vocational ] is 1.45 times the awar t rate for
    the general holding the other variables at constant. The percent change
    in the award rate of num_awards is an increase of 7% for every unit
    increase in math.
    Loglinear Models

    Consider an I × J contingency table that cross-classifies n subjects. When


    the responses are statistically independent, the joint cell probabilities{πij }
    are determined by the row and column marginal totals,
    πij = πi+π+j , i = 1,... ,I, j = 1,...,J

    Under independence, μij = nπi+π+j for all i and j .


    Loglinear Model of Independence for Two-Way
    Table
    ➔ Denote the row variable by X and the column variable by Y .
    The condition of independence, μij = nπi+π+j , is multiplicative.

    ➔ Thus, independence has the form logμij = λ + λXi + λY j

    for a row effect λXi and a column effect λY j


    .
    Loglinear Model of Independence for Two-Way Table
    Example
    Example: Belief in Life after Death
    For example, in the 2000 General Social Survey, subjects were asked whether they
    believed in life after death. The number who answered “yes” was 1339 of the 1639
    whites, 260 of the 315 blacks and 88 of the 110 classified as “other” on race. Table
    shows results of fitting the independence loglinear model to the 3 × 2 table. The
    model fits well. For the constraints used, λY1 = 1.50 and λY2 = 0. Therefore, the
    estimated odds of belief in life after death was exp(1.50) = 4.5 for each race

    Belief in Life after Death


    Race Yes No Total
    White 1339 300 1639
    1339.63 299.37
    Black 260 55 315
    257.46 57.536
    Others 88 22 110
    89.91 20.092
    Total 1687 377 2064
    Saturated Model for Two-Way Tables

    Variables that are statistically dependent rather than independent satisfy


    the more complex loglinear model,
    logμij = λ + λXi + λYj + λXYij

    The λXYij parameters are association terms that reflect deviations from
    independence.
    .
    Example: Belief in Life after Death

    The estimated odds ratios between


    belief and race are exp(0.1096) =
    1.12 for white and other, exp(0.1671)
    = 1.18 for black and other, and
    exp(0.1096 − 0.1671) = 0.94 for white
    and black. For example, the estimated
    odds of belief in an life after death for
    whites are 0.94 times the estimated
    odds for blacks. Since the
    independence model fitted well, none
    of these estimated odds ratios differ
    significantly from 1.0.

    You might also like