DOCSLIB.ORG

Randomization Does Not Justify Logistic Regression

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Randomization Does Not Justify Logistic Regression Statistical Science 2008, Vol. 23, No. 2, 237–249 DOI: 10.1214/08-STS262 c Institute of Mathematical Statistics, 2008 Randomization Does Not Justify Logistic Regression David A. Freedman Abstract. The logit model is often used to analyze experimental data. However, randomization does not justify the model, so the usual esti- mators can be inconsistent. A consistent estimator is proposed. Ney- man’s non-parametric setup is used as a benchmark. In this setup, each subject has two potential responses, one if treated and the other if untreated; only one of the two responses can be observed. Beside the mathematics, there are simulation results, a brief review of the literature, and some recommendations for practice. Key words and phrases: Models, randomization, logistic regression, logit, average predicted probability. 1. INTRODUCTION nπT subjects at random and assign them to the treatment condition. The remaining nπ subjects The logit model is often fitted to experimental C are assigned to a control condition, where π = 1 data. As explained below, randomization does not C π . According to Neyman (1923), each subject has− justify the assumptions behind the model. Thus, the T two responses: Y T if assigned to treatment, and Y C conventional estimator of log odds is difficult to in- i i if assigned to control. The responses are 1 or 0, terpret; an alternative will be suggested. Neyman’s where 1 is “success” and 0 is “failure.” Responses setup is used to define parameters and prove re- are fixed, that is, not random. sults. (Grammatical niceties apart, the terms “logit If i is assigned to treatment (T ), then Y T is ob- model” and “logistic regression” are used interchange- i served. Conversely, if i is assigned to control (C), ably.) then Y C is observed. Either one of the responses After explaining the models and estimators, we i may be observed, but not both. Thus, responses are present simulations to illustrate the findings. A brief subject-level parameters. Even so, responses are es- review of the literature describes the history and timable (see Section 9). Each subject has a covari- arXiv:0808.3914v1 [stat.ME] 28 Aug 2008 current usage. Some practical recommendations are ate Z , unaffected by assignment; Z is observable. derived from the theory. Analytic proofs are sketched i i In this setup, the only stochastic element is the ran- at the end of the paper. domization: conditional on the assignment variable X , the observed response Y = X Y T + (1 X )Y C 2. NEYMAN i i i i i i is deterministic. − There is a study population with n subjects in- Population-level ITT (intention-to-treat) param- dexed by i = 1,...,n. Fix πT with 0 <πT < 1. Choose eters are defined by taking averages over all n sub- jects in the study population: David A. Freedman is Professor, Department of 1 αT = Y T , Statistics, University of California, Berkeley, California n i 94720-3860, USA e-mail: [email protected]. (1) X 1 αC = Y C . This is an electronic reprint of the original article n i published by the Institute of Mathematical Statistics in T X Statistical Science, 2008, Vol. 23, No. 2, 237–249. This For example, α is the fraction of successes if all reprint differs from the original in pagination and subjects are assigned to T ; similarly for αC . A pa- typographic detail. rameter of considerable interest is the differential log 1 2 D. A. FREEDMAN odds of success, (To verify this, check first that Ui is distributed − αT αC like +Ui.) The parameter vector β = (β1, β2, β3) is (2) ∆ = log log . usually estimated by maximum likelihood. We de- 1 αT − 1 αC − − note the MLE by βˆ. The logit model is all about log odds (more on this below). The parameter ∆ defined by (2) may there- Interpreting the Coefficients in the Model fore be what investigators think is estimated by run- In the case of primary interest, Xi is 1 or 0. Con- T ning logistic regressions on experimental data, al- sider the log odds λi of success when Xi = 1, as well C though that idea is seldom explicit. as the log odds λi when Xi = 0. In view of (4), The Intention-to-Treat Principle T p(β, 1, Zi) λi = log 1 p(β, 1, Zi) The intention-to-treat principle, which goes back − to Hill (1961, page 259), is to make comparisons = β1 + β2 + β3Zi, based on treatment assigned rather than treatment (5) received. Such comparisons take full advantage of C p(β, 0, Zi) λi = log the randomization, thereby avoiding biases due to 1 p(β, 0, Zi) − self-selection. For example, the unbiased estimators = β1 + β3Zi. for the parameters in (1) are the fraction of successes T C In particular, λ λ = β2 for all i, whatever the in the treatment group and the control group, re- i − i spectively. Below, these will be called ITT estima- value of Zi may be. Thus, according to the model, tors. ITT estimators measure the effect of assign- Xi =1 adds β2 to the log odds of success. ment rather than treatment. With crossover, the Application to Experimental Data distinction matters. For additional discussion, see To apply the model to experimental data, define Freedman (2006a). Xi =1if i is assigned to T , while Xi =0if i assigned 3. THE LOGIT MODEL to C. Notice that the model not justified by random- ization. Why would the logit specification be correct To set up the logit model, we consider a study rather than the probit—or anything else? What jus- population of n subjects, indexed by i = 1,...,n. tifies the choice of covariates? Why are they exoge- Each subject has three observable random variables: nous? If the model is wrong, what is βˆ2 supposed Yi, Xi, Zi. Here, Yi is the response, which is 0 or 1. to be estimating? The last rhetorical question may The primary interest is the “effect” of Xi on Yi, and have an answer: the parameter ∆ in (2) seems like Zi is a covariate. a natural choice, as indicated above. For our purposes, the best way to formulate the More technically, from Neyman’s perspective, given model involves a latent (unobservable) random vari- the assignment variables Xi , the responses are de- T { } C able Ui for each subject. These are assumed to be terministic: Yi = Yi if Xi = 1, while Yi = Yi if Xi = independent across subjects, with a common logistic 0. The logit model, on the other hand, views the re- distribution: for <u< , sponses Y as random—with a specified distribution— −∞ ∞ i given the{ assignment} variables and covariates. (3) P (Ui < u) = exp(u)/[1 + exp(u)], u The contrast is therefore between two styles of where exp(u)= e . The model assumes that X and inference. Z are exogenous, that is, independent of U. More formally, Xi, Zi : i = 1,...,n is assumed to inde- Randomization provides a known distribution for { } • pendent of Ui : i = 1,...,n . Finally, the model as- the assignment variables; statistical inferences are { } sumes that Yi =1 if based on this distribution. Modeling assumes a distribution for the latent β1 + β2Xi + β3Zi + Ui > 0; • variables; statistical inferences are based on that else, Yi = 0. assumption. Furthermore, model-based inferences Given X and Z, it follows that responses are inde- are conditional on the assignment variables and pendent across subjects, the conditional probability covariates. that Yi = 1 being p(β, Xi, Zi), where A similar contrast will be found in other areas too, exp(β1 + β2x + β3z) including sample surveys. See Koch and Gillings (2005) (4) p(β,x,z)= . 1 + exp(β1 + β2x + β3z) for a review and pointers to the literature. LOGISTIC REGRESSION 3 What if the Logit Model is Right? From Neyman to Logits Suppose the model is right, and there is a causal How could we get from Neyman to the logit model? T C interpretation. We can intervene and set Xi to 1 To begin with, we would allow Yi and Yi to be 0–1 without changing the Z’s or U’s, so Yi =1 if and valued random variables; the Zi can be random too. only if β1 + β2 + β3Zi + Ui > 0. Similarly, we can set To define the parameters in (1) and (2), we would re- T C Xi to 0 without changing anything else, and then place Yi and Yi by their expectations. None of this Yi = 1 if and only if β1 + β3Zi + Ui > 0. Notice that is problematic, and the Neyman model is now ex- β2 appears when Xi is set to 1, but disappears when tremely general and flexible. Randomization makes Xi is set to 0. the assignment variables Xi independent of the T {C } On this basis, for each subject, whatever the value potential responses Yi ,Yi . of Zi may be, setting Xi to 1 rather than 0 adds β2 To get the logit model, however, we would need to the log odds of success. If the model is right, β2 is to specialize this setup considerably, assuming the a very useful parameter, which is well estimated by existence of IID logistic random variables Ui, inde- the MLE provided n is large. For additional detail pendent of the covariates Zi, with T on causal modeling and estimation, see Freedman Yi = 1 if and only if (2005). β1 + β2 + β3Zi + Ui > 0, Even if the model is right and n is large, β2 differs (9) T C from ∆ in (2). For instance, α will be nearly equal Yi = 1 if and only if 1 n T T to n i=1 p(β, 1, Zi).
Load more

Recommended publications
  • Logistic Regression, Dependencies, Non-Linear Data and Model Reduction
    COMP6237 – Logistic Regression, Dependencies, Non-linear Data and Model Reduction Markus Brede [email protected] Lecture slides available here: http://users.ecs.soton.ac.uk/mb8/stats/datamining.html (Thanks to Jason Noble and Cosma Shalizi whose lecture materials I used to prepare) COMP6237: Logistic Regression ● Outline: – Introduction – Basic ideas of logistic regression – Logistic regression using R – Some underlying maths and MLE – The multinomial case – How to deal with non-linear data ● Model reduction and AIC – How to deal with dependent data – Summary – Problems Introduction ● Previous lecture: Linear regression – tried to predict a continuous variable from variation in another continuous variable (E.g. basketball ability from height) ● Here: Logistic regression – Try to predict results of a binary (or categorical) outcome variable Y from a predictor variable X – This is a classification problem: classify X as belonging to one of two classes – Occurs quite often in science … e.g. medical trials (will a patient live or die dependent on medication?) Dependent variable Y Predictor Variables X The Oscars Example ● A fictional data set that looks at what it takes for a movie to win an Oscar ● Outcome variable: Oscar win, yes or no? ● Predictor variables: – Box office takings in millions of dollars – Budget in millions of dollars – Country of origin: US, UK, Europe, India, other – Critical reception (scores 0 … 100) – Length of film in minutes – This (fictitious) data set is available here: https://www.southampton.ac.uk/~mb1a10/stats/filmData.txt Predicting Oscar Success ● Let's start simple and look at only one of the predictor variables ● Do big box office takings make Oscar success more likely? ● Could use same techniques as below to look at budget size, film length, etc.
    [Show full text]
  • Logistic Regression Maths and Statistics Help Centre
    Logistic regression Maths and Statistics Help Centre Many statistical tests require the dependent (response) variable to be continuous so a different set of tests are needed when the dependent variable is categorical. One of the most commonly used tests for categorical variables is the Chi-squared test which looks at whether or not there is a relationship between two categorical variables but this doesn’t make an allowance for the potential influence of other explanatory variables on that relationship. For continuous outcome variables, Multiple regression can be used for a) controlling for other explanatory variables when assessing relationships between a dependent variable and several independent variables b) predicting outcomes of a dependent variable using a linear combination of explanatory (independent) variables The maths: For multiple regression a model of the following form can be used to predict the value of a response variable y using the values of a number of explanatory variables: y 0 1x1 2 x2 ..... q xq 0 Constant/ intercept , 1 q co efficients for q explanatory variables x1 xq The regression process finds the co-efficients which minimise the squared differences between the observed and expected values of y (the residuals). As the outcome of logistic regression is binary, y needs to be transformed so that the regression process can be used. The logit transformation gives the following: p ln 0 1x1 2 x2 ..... q xq 1 p p p probabilty of event occuring e.g. person dies following heart attack, odds ratio 1- p If probabilities of the event of interest happening for individuals are needed, the logistic regression equation exp x x ....
    [Show full text]
  • Generalized Linear Models
    CHAPTER 6 Generalized linear models 6.1 Introduction Generalized linear modeling is a framework for statistical analysis that includes linear and logistic regression as special cases. Linear regression directly predicts continuous data y from a linear predictor Xβ = β0 + X1β1 + + Xkβk.Logistic regression predicts Pr(y =1)forbinarydatafromalinearpredictorwithaninverse-··· logit transformation. A generalized linear model involves: 1. A data vector y =(y1,...,yn) 2. Predictors X and coefficients β,formingalinearpredictorXβ 1 3. A link function g,yieldingavectoroftransformeddataˆy = g− (Xβ)thatare used to model the data 4. A data distribution, p(y yˆ) | 5. Possibly other parameters, such as variances, overdispersions, and cutpoints, involved in the predictors, link function, and data distribution. The options in a generalized linear model are the transformation g and the data distribution p. In linear regression,thetransformationistheidentity(thatis,g(u) u)and • the data distribution is normal, with standard deviation σ estimated from≡ data. 1 1 In logistic regression,thetransformationistheinverse-logit,g− (u)=logit− (u) • (see Figure 5.2a on page 80) and the data distribution is defined by the proba- bility for binary data: Pr(y =1)=y ˆ. This chapter discusses several other classes of generalized linear model, which we list here for convenience: The Poisson model (Section 6.2) is used for count data; that is, where each • data point yi can equal 0, 1, 2, ....Theusualtransformationg used here is the logarithmic, so that g(u)=exp(u)transformsacontinuouslinearpredictorXiβ to a positivey ˆi.ThedatadistributionisPoisson. It is usually a good idea to add a parameter to this model to capture overdis- persion,thatis,variationinthedatabeyondwhatwouldbepredictedfromthe Poisson distribution alone.
    [Show full text]
  • An Introduction to Logistic Regression: from Basic Concepts to Interpretation with Particular Attention to Nursing Domain
    J Korean Acad Nurs Vol.43 No.2, 154 -164 J Korean Acad Nurs Vol.43 No.2 April 2013 http://dx.doi.org/10.4040/jkan.2013.43.2.154 An Introduction to Logistic Regression: From Basic Concepts to Interpretation with Particular Attention to Nursing Domain Park, Hyeoun-Ae College of Nursing and System Biomedical Informatics National Core Research Center, Seoul National University, Seoul, Korea Purpose: The purpose of this article is twofold: 1) introducing logistic regression (LR), a multivariable method for modeling the relationship between multiple independent variables and a categorical dependent variable, and 2) examining use and reporting of LR in the nursing literature. Methods: Text books on LR and research articles employing LR as main statistical analysis were reviewed. Twenty-three articles published between 2010 and 2011 in the Journal of Korean Academy of Nursing were analyzed for proper use and reporting of LR models. Results: Logistic regression from basic concepts such as odds, odds ratio, logit transformation and logistic curve, assumption, fitting, reporting and interpreting to cautions were presented. Substantial short- comings were found in both use of LR and reporting of results. For many studies, sample size was not sufficiently large to call into question the accuracy of the regression model. Additionally, only one study reported validation analysis. Conclusion: Nurs- ing researchers need to pay greater attention to guidelines concerning the use and reporting of LR models. Key words: Logit function, Maximum likelihood estimation, Odds, Odds ratio, Wald test INTRODUCTION The model serves two purposes: (1) it can predict the value of the depen- dent variable for new values of the independent variables, and (2) it can Multivariable methods of statistical analysis commonly appear in help describe the relative contribution of each independent variable to general health science literature (Bagley, White, & Golomb, 2001).
    [Show full text]
  • Variance Partitioning in Multilevel Logistic Models That Exhibit Overdispersion
    J. R. Statist. Soc. A (2005) 168, Part 3, pp. 599–613 Variance partitioning in multilevel logistic models that exhibit overdispersion W. J. Browne, University of Nottingham, UK S. V. Subramanian, Harvard School of Public Health, Boston, USA K. Jones University of Bristol, UK and H. Goldstein Institute of Education, London, UK [Received July 2002. Final revision September 2004] Summary. A common application of multilevel models is to apportion the variance in the response according to the different levels of the data.Whereas partitioning variances is straight- forward in models with a continuous response variable with a normal error distribution at each level, the extension of this partitioning to models with binary responses or to proportions or counts is less obvious. We describe methodology due to Goldstein and co-workers for appor- tioning variance that is attributable to higher levels in multilevel binomial logistic models. This partitioning they referred to as the variance partition coefficient. We consider extending the vari- ance partition coefficient concept to data sets when the response is a proportion and where the binomial assumption may not be appropriate owing to overdispersion in the response variable. Using the literacy data from the 1991 Indian census we estimate simple and complex variance partition coefficients at multiple levels of geography in models with significant overdispersion and thereby establish the relative importance of different geographic levels that influence edu- cational disparities in India. Keywords: Contextual variation; Illiteracy; India; Multilevel modelling; Multiple spatial levels; Overdispersion; Variance partition coefficient 1. Introduction Multilevel regression models (Goldstein, 2003; Bryk and Raudenbush, 1992) are increasingly being applied in many areas of quantitative research.
    [Show full text]
  • Chapter 19: Logistic Regression
    DISCOVERING STATISTICS USING SPSS Chapter 19: Logistic regression Smart Alex’s Solutions Task 1 A ‘display rule’ refers to displaying an appropriate emotion in a given situation. For example, if you receive a Christmas present that you don’t like, the appropriate emotional display is to smile politely and say ‘Thank you Why did you buy me this Auntie Kate, I’ve always wanted a rotting cabbage’. The crappy statistics textbook for Christmas, Auntie Kate? inappropriate emotional display is to start crying and scream ‘Why did you buy me a rotting cabbage, you selfish old bag?’ A psychologist measured children’s understanding of display rules (with a task that they could either pass or fail), their age (months), and their ability to understand others’ mental states (‘theory of mind’, measured with a false-belief task that they could pass or fail). The data are in Display.sav. Can display rule understanding (did the child pass the test: yes/no?) be predicted from the theory of mind (did the child pass the false-belief task: yes/no?), age and their interaction? The main analysis To carry out logistic regression, the data must be entered as for normal regression: they are arranged in the data editor in three columns (one representing each variable). Open the file Display.sav. Looking at the data editor, you should notice that both of the categorical variables have been entered as coding variables; that is, numbers have been specified to represent categories. For ease of interpretation, the outcome variable should be coded 1 (event occurred) and 0 (event did not occur); in this case, 1 represents having display rule understanding, and 0 represents an absence of display rule understanding.
    [Show full text]
  • An Introduction to Biostatistics: Part 2
    An introduction to biostatistics: part 2 Cavan Reilly September 4, 2019 Table of contents Statistical models Maximum likelihood estimation Linear models Multiple regression Confounding Statistical adjustments Contrasts Interactions Generalized linear models Logistic regression Model selection Statistical models Statistical models can be powerful tools for understanding complex relationships among variables. We will first suppose that we observe 2 continuous variables. Typically we would start out by looking at a scatterplot: so let's look at an example. Let's read in some data from the genetic association study we looked at in the previous lecture: fms=read.delim("http://www.biostat.umn.edu/~cavanr/FMS_data.txt") Statistical models Then we can just use the plot command as follows. > pdf("wght-hght.pdf") > plot(fms$Pre.height,fms$Pre.weight,xlab="Height",ylab="Weight") > dev.off() It looks like the 2 variables increase together, but we clearly don't have an equation like: Weight = β0 + β1Height; for 2 constants β0 and β1. Note: we look at the distribution of the response variable conditional on the explanatory variable. Correlation A commonly used measure of the extent to which 2 continuous variables have a linear association is the correlation coefficient. The cor function in R allows one to compute this summary. If it is positive then large values of one variable are generally associated with large values of the other. If it is negative then large values of one variables are associated with small values of the other. If the absolute value of the correlation coefficient exceeds 0.7 then there is a strong association, if less than 0.3 then weak, otherwise moderate.
    [Show full text]
  • Lecture 9: Logistic Regression (V2)
    MS&E 226: \Small" Data Lecture 9: Logistic regression (v2) Ramesh Johari [email protected] 1 / 28 Regression methods for binary outcomes 2 / 28 Binary outcomes For the duration of this lecture suppose the outcome variable Yi 2 f0; 1g for each i. (Much of what we cover generalizes to discrete outcomes with more than two levels, but we focus on the binary case.) 3 / 28 Linear regression? Why doesn't linear regression work well for prediction? A picture: 1 Y 0 −20 0 20 X 4 / 28 Logistic regression At its core, logistic regression is a method that directly addresses this issue with linear regression: it produces fitted values that always lie in [0; 1]. Input: Sample data X and Y. Output: A fitted model f^(·), where we interpret f^(X~ ) as an estimate of the probability that the corresponding outcome Y is equal to 1. 5 / 28 Logistic regression: Basics 6 / 28 The population model Logistic regression assumes the following population model: exp(X~ β) P(Y = 1jX~ ) = = 1 − P(Y = 0jX~ ): 1 + exp(X~ β) A plot in the case with only one covariate, and β0 = 0; β1 = 1: 1.00 0.75 0.50 P(Y = 1 | X) 0.25 0.00 −10 −5 0 5 10 X 7 / 28 Logistic curve The function g given by: q g(q) = log 1 − q is called the logit function. It has the following inverse, called the logistic curve: exp(z) g−1(z) = : 1 + exp(z) In terms of g, we can write the population model as:1 −1 P(Y = 1jX~ ) = g (X~ β): 1This is one example of a generalized linear model (GLM); for a GLM, g is called the link function.
    [Show full text]
  • Logistic Regression, Part I: Problems with the Linear Probability Model
    Logistic Regression, Part I: Problems with the Linear Probability Model (LPM) Richard Williams, University of Notre Dame, https://www3.nd.edu/~rwilliam/ Last revised February 22, 2015 This handout steals heavily from Linear probability, logit, and probit models, by John Aldrich and Forrest Nelson, paper # 45 in the Sage series on Quantitative Applications in the Social Sciences. INTRODUCTION. We are often interested in qualitative dependent variables: • Voting (does or does not vote) • Marital status (married or not) • Fertility (have children or not) • Immigration attitudes (opposes immigration or supports it) In the next few handouts, we will examine different techniques for analyzing qualitative dependent variables; in particular, dichotomous dependent variables. We will first examine the problems with using OLS, and then present logistic regression as a more desirable alternative. OLS AND DICHOTOMOUS DEPENDENT VARIABLES. While estimates derived from regression analysis may be robust against violations of some assumptions, other assumptions are crucial, and violations of them can lead to unreasonable estimates. Such is often the case when the dependent variable is a qualitative measure rather than a continuous, interval measure. If OLS Regression is done with a qualitative dependent variable • it may seriously misestimate the magnitude of the effects of IVs • all of the standard statistical inferences (e.g. hypothesis tests, construction of confidence intervals) are unjustified • regression estimates will be highly sensitive to the range of particular values observed (thus making extrapolations or forecasts beyond the range of the data especially unjustified) OLS REGRESSION AND THE LINEAR PROBABILITY MODEL (LPM). The regression model places no restrictions on the values that the independent variables take on.
    [Show full text]
  • Generalized Linear Models Link Function the Logistic Equation Is
    Newsom Psy 525/625 Categorical Data Analysis, Spring 2021 1 Generalized Linear Models Link Function The logistic equation is stated in terms of the probability that Y = 1, which is π, and the probability that Y = 0, which is 1 - π. π ln =αβ + X 1−π The left-hand side of the equation represents the logit transformation, which takes the natural log of the ratio of the probability that Y is equal to 1 compared to the probability that it is not equal to one. As we know, the probability, π, is just the mean of the Y values, assuming 0,1 coding, which is often expressed as µ. The logit transformation could then be written in terms of the mean rather than the probability, µ ln =αβ + X 1− µ The transformation of the mean represents a link to the central tendency of the distribution, sometimes called the location, one of the important defining aspects of any given probability distribution. The log transformation represents a kind of link function (often canonical link function)1 that is sometimes given more generally as g(.), with the letter g used as an arbitrary name for a mathematical function and the use of the “.” within the parentheses to suggest that any variable, value, or function (the argument) could be placed within. For logistic regression, this is known as the logit link function. The right hand side of the equation, α + βX, is the familiar equation for the regression line and represents a linear combination of the parameters for the regression. The concept of this logistic link function can generalized to any other distribution, with the simplest, most familiar case being the ordinary least squares or linear regression model.
    [Show full text]
  • Binary Logistic Regression
    Binary Logistic Regression The coefficients of the multiple regression model are estimated using sample data with k independent variables Estimated Estimated Estimated slope coefficients (or predicted) intercept value of Y ˆ Yi = b0 + b1 X 1i + b2 X 2i ++ bk X ki • Interpretation of the Slopes: (referred to as a Net Regression Coefficient) – b1=The change in the mean of Y per unit change in X1, taking into account the effect of X2 (or net of X2) 2 – b0 Y intercept. It is the same as simple regression. Binary Logistic Regression • Binary logistic regression is a type of regression analysis where the dependent variable is a dummy variable (coded 0, 1) • Why not just use ordinary least squares? ^ Y = a + bx – You would typically get the correct answers in terms of the sign and significance of coefficients – However, there are three problems 3 Binary Logistic Regression OLS on a dichotomous dependent variable: Yes = 1 No = 0 Y = Support Privatizing Social Security 1 10 X = Income 4 Binary Logistic Regression – However, there are three problems 1. The error terms are heteroskedastic (variance of the dependent variable is different with different values of the independent variables 2. The error terms are not normally distributed 3. And most importantly, for purpose of interpretation, the predicted probabilities can be greater than 1 or less than 0, which can be a problem for subsequent analysis. 5 Binary Logistic Regression • The “logit” model solves these problems: – ln[p/(1-p)] = a + BX or – p/(1-p) = ea + BX – p/(1-p) = ea (eB)X Where: “ln” is the natural logarithm, logexp, where e=2.71828 “p” is the probability that Y for cases equals 1, p (Y=1) “1-p” is the probability that Y for cases equals 0, 1 – p(Y=1) “p/(1-p)” is the odds ln[p/1-p] is the log odds, or “logit” 6 Binary Logistic Regression • Logistic Distribution P (Y=1) x • Transformed, however, the “log odds” are linear.
    [Show full text]
  • Bayesian Logistic Regression, Bayesian Generative Classification
    Bayesian Logistic Regression, Bayesian Generative Classification Piyush Rai Topics in Probabilistic Modeling and Inference (CS698X) Jan 23, 2019 Prob. Mod. & Inference - CS698X (Piyush Rai, IITK) Bayesian Logistic Regression, Bayesian Generative Classification 1 Bayesian Logistic Regression Recall that the likelihood model for logistic regression is Bernoulli (since y 2 f0; 1g) " #y " #(1−y) exp(w >x) 1 p(yjx; w) = Bernoulli(σ(w >x)) = = µy (1 − µ)1−y 1 + exp(w >x) 1 + exp(w >x) Just like the Bayesian linear regression case, let's use a Gausian prior on w λ p(w) = N (0; λ−1I ) / exp(− w >w) D 2 N Given N observations (X; y) = fx n; yngn=1, where X is N × D and y is N × 1, the posterior over w QN p(yjX; w)p(w) p(ynjx n; w)p(w) p(wjX; y) = = n=1 R p(yjX; w)p(w)dw R QN n=1 p(ynjx n; w)p(w)dw The denominator is intractable in general (logistic-Bernoulli and Gaussian are not conjugate) Can't get a closed form expression for p(wjX; y). Must approximate it! Several ways to do it, e.g., MCMC, variational inference, Laplace approximation (today) Prob. Mod. & Inference - CS698X (Piyush Rai, IITK) Bayesian Logistic Regression, Bayesian Generative Classification 2 Laplace Approximation of Posterior Distribution p(Djθ)p(θ) p(D,θ) Approximate the posterior distribution p(θjD) = p(D) = p(D) by the following Gaussian −1 p(θjD) ≈ N (θMAP ; H ) Note: θMAP is the maximum-a-posteriori (MAP) estimate of θ, i.e., θMAP = arg max p(θjD) = arg max p(D; θ) = arg max p(Djθ)p(θ) = arg max[log p(Djθ) + log p(θ)] θ θ θ θ Usually θMAP can be easily solved for (e.g., using first/second order iterative methods) H is the Hessian matrix of the negative log-posterior (or negative log-joint-prob) at θMAP 2 2 2 H = −∇ log p(θjD) = −∇ log p(D; θ) = −∇ [log p(Djθ) + log p(θ)] θ=θMAP θ=θMAP θ=θMAP Prob.
    [Show full text]