<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>vcovCL on Everyday Is A School Day</title>
    <link>https://www.kenkoonwong.com/tags/vcovcl/</link>
    <description>Recent content in vcovCL on Everyday Is A School Day</description>
    <generator>Hugo -- gohugo.io</generator>
    <language>en</language>
    <lastBuildDate>Fri, 20 Nov 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://www.kenkoonwong.com/tags/vcovcl/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Why Risk Ratio Is Preferred, What is Modified Poisson, and Why Do We Even Care About Standard Error?</title>
      <link>https://www.kenkoonwong.com/blog/modified-poisson/</link>
      <pubDate>Fri, 20 Nov 2026 00:00:00 +0000</pubDate>
      
      <guid>https://www.kenkoonwong.com/blog/modified-poisson/</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;Learnt about modified Poisson for the first time. Simulated RR vs OR, log-binomial and robust SE, then sprinkled in repeated measures. Turns out a 95% CI can be too narrow (confidently wrong) or too wide (overly cautious)! Still learning, but pretty convinced now 🙌&lt;/p&gt;
&lt;/blockquote&gt;




&lt;h2 id=&#34;motivations&#34;&gt;Motivations
  &lt;a href=&#34;#motivations&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;p&gt;I was working on a project recently and learnt a few things as we were going along and want to look into it a bit more with simulation to see for myself. These might sound like some random topics, but they&amp;rsquo;re all kind of tied together, at least in my mind they are. I&amp;rsquo;ve not heard of modified poisson before and what a great way to get from incompetent incompetence to at least competent incompetence. 🤣 Knowing what we don&amp;rsquo;t know is definitely better than not knowing what we don&amp;rsquo;t know (MNAR problem). Well let&amp;rsquo;s explore together and see for ourselves, will this be our next go-to model along with robust sandwich standard error? Oh, let&amp;rsquo;s also sprinkle a little bit of repeatd measure (some clustering) to beef up that SE to see an even better effect with this robust sandwich standard error. Let&amp;rsquo;s go! 🙌&lt;/p&gt;




&lt;h2 id=&#34;thought-process&#34;&gt;Thought Process
  &lt;a href=&#34;#thought-process&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;p&gt;But, how are we going to do this? Let&amp;rsquo;s go through the bare basics of the interpretation, let&amp;rsquo;s say we have binary outcome. A heuristic is to use logistic regression model with logit as link function. It&amp;rsquo;s a great model, but exponentiating on the coefficient of exposure turns it into odds ratio instead of risk ratio (RR). So, &lt;code&gt;let&#39;s talk a bit about why risk ratio may be a bit easier to understand&lt;/code&gt; than odds ratio. After this we&amp;rsquo;ll look and see how can we &lt;code&gt;simulate using baseline rate and RR&lt;/code&gt;. And use this simulated data to &lt;code&gt;check on our regular poisson regression, log binomial regression, logistic regression with logit link, then poisson regression w robust SE (aka modified poisson)&lt;/code&gt;. We&amp;rsquo;ll look at their wideness of 95%CI. Then, to sprinkle a little bit of sweetness on top, we&amp;rsquo;ll add on repeated measure, look at SE of all these models. Top it up with, g-computation using logistic regression &amp;amp; bootstrap based on subject and see how that matches up with modified poisson.&lt;/p&gt;




&lt;h2 id=&#34;objectives&#34;&gt;Objectives:
  &lt;a href=&#34;#objectives&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;a href=&#34;#rr&#34;&gt;Why Risk Ratio?&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#sim&#34;&gt;Simulate Baseline Rate and RR&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#comparison&#34;&gt;Comparison&lt;/a&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;a href=&#34;#poisson&#34;&gt;Poisson Regression&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#logbin&#34;&gt;Log Binomial Regression&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#logit&#34;&gt;Logistic Regression with Logit Link&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#modpoisson&#34;&gt;Poisson Regression with Robust Standard Error (Modified Poisson)&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#highevent&#34;&gt;High Event Rates&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#high-rr&#34;&gt;High RR&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#very-high-rr&#34;&gt;Very High RR&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#repeated&#34;&gt;Repeat Measures&lt;/a&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;a href=&#34;#gcomp&#34;&gt;RR Without Modified Poisson with LR G-computation with subject level bootstrap&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#replicate-of-sim&#34;&gt;Replicates of Our 4 Scenarios&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#opportunities&#34;&gt;Opportunities For Improvement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;
&lt;a href=&#34;#lessons&#34;&gt;Lessons Learnt&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;




&lt;h2 id=&#34;rr&#34;&gt;Why Risk Ratio?
  &lt;a href=&#34;#rr&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;p&gt;Risk Ratio is easier to understand because it&amp;rsquo;s basically risk of event in exposed group divided by risk of event in unexposed group (p1/p0). So, if we have an RR of 0.6, we&amp;rsquo;re saying the exposed group has 40% less risk of event than the unexposed group. If we have an RR of 1.5, we&amp;rsquo;re saying the exposed group has 50% more risk of event than the unexposed group. That’s much more intuitive than the odds ratio, which only approximates RR when the outcome is rare (roughly &amp;lt;10%). When the outcome is common, OR exaggerates the effect, pulling further from 1 than the RR.&lt;/p&gt;
&lt;p&gt;To get an RR, we can’t just exponentiate the exposure coefficient from a logistic regression, because that gives us an OR. Instead, we can use:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;Log-binomial regression&lt;/code&gt; (binomial family, log link): this models RR directly, but it often fails to converge, especially with continuous covariates or high baseline risk.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Modified Poisson regression&lt;/code&gt; (Poisson family, log link, robust SEs): exponentiating the exposure coefficient gives us RR, how convenient! But there’s a catch. With a binary outcome, plain Poisson overestimates the variance, so the CIs are too wide. Use robust (sandwich) standard errors to fix this, e.g. sandwich::vcovHC() in R. Predicted “risks” can also exceed 1, so don’t use the model for predictions.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Another reason to use RR to communicate effect size is that OR is non-collapsible. What do that mean? 🤷‍♂️ I&amp;rsquo;ve tried to understand this several times in the past, I thought I got it but it was an illusion of understanding. 🤣 We&amp;rsquo;ll have to learn and go deeper on what that means next time with simulations and examples. Let&amp;rsquo;s focus on our current task.&lt;/p&gt;




&lt;h2 id=&#34;sim&#34;&gt;Simulate Baseline Rate and RR
  &lt;a href=&#34;#sim&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;library&lt;/span&gt;(tidyverse)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;library&lt;/span&gt;(broom)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;library&lt;/span&gt;(sandwich)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;library&lt;/span&gt;(lmtest)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;set.seed&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;n &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#099&#34;&gt;1300&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;baseline &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;log&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;0.06&lt;/span&gt;) &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# 6% base rate&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;rr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;log&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;0.6&lt;/span&gt;) &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# RR of 0.6, meaning treatment reduces risk for 40% compared to non-treatment (noticed how we switched the term exposure with treatment here)&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rbinom&lt;/span&gt;(n,&lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;,&lt;span style=&#34;color:#099&#34;&gt;0.5&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;p &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;exp&lt;/span&gt;(baseline &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt; rr&lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;*&lt;/span&gt;x)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rbinom&lt;/span&gt;(n, &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;, p)
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;We&amp;rsquo;ve not simulated with the above setting before. So, let&amp;rsquo;s try to explain so that our future self can return and recall what&amp;rsquo;s going on here.&lt;/p&gt;
&lt;p&gt;So in Poisson regression, we have log link function. So, we can model the log of the expected count as a linear combination of the predictors. In this case, we have a binary predictor x (0 or 1) and we want to simulate a binary outcome y (0 or 1) based on a baseline risk and a risk ratio. Poisson formula is log(λ) = β0 + β1*x, where λ is the expected count (or rate), β0 is the intercept (baseline log risk), and β1 is the coefficient for x (log risk ratio). To get probability of the Poisson distribution, we can exponentiate the linear predictor to get λ = exp(β0 + β1*x). Then we can use this λ as the probability of success in a Bernoulli trial to simulate our binary outcome y. Later on, we can add confounder to it on the equation as well.&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;table&lt;/span&gt;(y)
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## y
##    0    1 
## 1228   72
&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;Pretty small event isn&amp;rsquo;t it? Let&amp;rsquo;s start with this&lt;/p&gt;




&lt;h2 id=&#34;comparison&#34;&gt;Comparison
  &lt;a href=&#34;#comparison&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;




&lt;h3 id=&#34;poisson&#34;&gt;Poisson Regression
  &lt;a href=&#34;#poisson&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;model_poisson &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;glm&lt;/span&gt;(y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; x, family &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; poisson)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;(model_poisson_df &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(model_poisson, exponentiate &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;, conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 2 × 7
##   term        estimate std.error statistic  p.value conf.low conf.high
##   &amp;lt;chr&amp;gt;          &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;
## 1 (Intercept)   0.0675     0.147    -18.3  1.24e-74   0.0499    0.0890
## 2 x             0.622      0.245     -1.94 5.28e- 2   0.379     0.997
&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;Remember, Poisson assumes that the mean is equal to the variance. But with binary outcome, the variance is p*(1-p), which is less than the mean (p). So, Poisson overestimates the variance, leading to wider confidence intervals. Also notice that the exponentiated coef of x is very close to our truth &lt;code&gt;0.6&lt;/code&gt;? 🙌 Wow! I would never have thought we can use Poisson model for binary data! Let&amp;rsquo;s keep an eye on the 95%CI width, it&amp;rsquo;s 0.6180093&lt;/p&gt;




&lt;h3 id=&#34;logbin&#34;&gt;Log Binomial Regression
  &lt;a href=&#34;#logbin&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;model_bin &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tryCatch&lt;/span&gt;(
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;glm&lt;/span&gt;(y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; x, family &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;binomial&lt;/span&gt;(&lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;log&amp;#34;&lt;/span&gt;)), &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;#notice the link function here is log, not logit?&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  error &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;function&lt;/span&gt;(e) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;NULL&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;(model_bin_df &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(model_bin, exponentiate &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;, conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 2 × 7
##   term        estimate std.error statistic  p.value conf.low conf.high
##   &amp;lt;chr&amp;gt;          &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;
## 1 (Intercept)   0.0675     0.142    -18.9  6.63e-80   0.0503    0.0880
## 2 x             0.622      0.239     -1.99 4.68e- 2   0.384     0.985
&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;wow, same estimate! But smalled 95% CI, which is because of smaller SE! The width of 95%CI is 0.6007632. Now let&amp;rsquo;s take a look at logistic regression.&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;Now one might wonder, if log binomial regression is the OG, why don&amp;rsquo;t we use this? We can, but it has a convergence issue, especially when we have continuous covariates or high baseline risk. That is also why we sometimes have to use &lt;code&gt;tryCatch&lt;/code&gt; for the error.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;h3 id=&#34;logit&#34;&gt;Logistic Regression with Logit Link
  &lt;a href=&#34;#logit&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;model_logit &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;glm&lt;/span&gt;(y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; x, family &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;binomial&lt;/span&gt;(&lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;logit&amp;#34;&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;(model_logit_df &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(model_logit, exponentiate &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;, conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 2 × 7
##   term        estimate std.error statistic  p.value conf.low conf.high
##   &amp;lt;chr&amp;gt;          &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;
## 1 (Intercept)   0.0724     0.153    -17.2  3.06e-66   0.0530    0.0965
## 2 x             0.605      0.252     -1.99 4.62e- 2   0.365     0.984
&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;OK, remember we mentioned early? the exponentiated coefficient does not give an RR, it&amp;rsquo;s OR, though very close to our truth, because the event is rare. But the width of 95%CI is 0.6189336. So, we can see that the logistic regression has a wider CI than the log-binomial regression, which is expected because the log-binomial model is more efficient for estimating RR when the outcome is binary.&lt;/p&gt;




&lt;h3 id=&#34;modpoisson&#34;&gt;Poisson Regression with Robust Standard Error (Modified Poisson)
  &lt;a href=&#34;#modpoisson&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;(model_modpoisson &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;coeftest&lt;/span&gt;(model_poisson, vcov &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;vcovHC&lt;/span&gt;(model_poisson, type &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;HC0&amp;#34;&lt;/span&gt;)) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;TRUE&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;#apparently exp not supported in coeftest from broom&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;across&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;c&lt;/span&gt;(estimate, conf.low, conf.high), exp)))
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 2 × 7
##   term        estimate std.error statistic  p.value conf.low conf.high
##   &amp;lt;chr&amp;gt;          &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;
## 1 (Intercept)   0.0675     0.142    -18.9  6.67e-80   0.0511    0.0893
## 2 x             0.622      0.239     -1.99 4.68e- 2   0.389     0.993
&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;Alright, accurate estimate (well, we didn&amp;rsquo;t change it from Poisson result), and smaller CI width than Poisson regression, but still wider than log-binomial regression. The width of 95%CI is 0.6040859 by very small amount! And doesn&amp;rsquo;t seem like there is convergence issue with modified poisson, which is good. 👍 We can read more about this here 
&lt;a href=&#34;https://academic.oup.com/aje/article-abstract/159/7/702/71883?login=false&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;Zou G - A Modified Poisson Regression Approach to Prospective Studies with Binary Data&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Let&amp;rsquo;s compare all of them with dataviz
&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-8-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;&lt;/p&gt;




&lt;h3 id=&#34;highevent&#34;&gt;High Event Rates
  &lt;a href=&#34;#highevent&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;p&gt;Let&amp;rsquo;s set baseline rate to 80%&lt;/p&gt;
&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## y
##   0   1 
## 441 859
&lt;/code&gt;&lt;/pre&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 4 × 5
##   model            estimate conf.low conf.high width
##   &amp;lt;fct&amp;gt;               &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt; &amp;lt;dbl&amp;gt;
## 1 Poisson             0.631    0.548     0.724 0.176
## 2 Log Binomial        0.631    0.577     0.686 0.109
## 3 Logistic            0.253    0.197     0.323 0.125
## 4 Modified Poisson    0.631    0.578     0.688 0.109
&lt;/code&gt;&lt;/pre&gt;&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-9-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;
&lt;p&gt;Wow, how about that! OR no longer is interchangeable as RR, 95% CI is wider for Poisson but for log binomial and modified Poisson, the 95% CIs are narrower and close to each other. Let&amp;rsquo;s increass the RR and see if log binomial breaks&lt;/p&gt;




&lt;h3 id=&#34;high-rr&#34;&gt;High RR
  &lt;a href=&#34;#high-rr&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;p&gt;Let&amp;rsquo;s set RR as 2 and base rate as 30%&lt;/p&gt;
&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## y
##   0   1 
## 749 551
&lt;/code&gt;&lt;/pre&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 4 × 5
##   model            estimate conf.low conf.high width
##   &amp;lt;fct&amp;gt;               &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt; &amp;lt;dbl&amp;gt;
## 1 Poisson              2.02     1.70      2.42 0.713
## 2 Log Binomial         2.02     1.77      2.33 0.558
## 3 Logistic             3.42     2.72      4.31 1.59 
## 4 Modified Poisson     2.02     1.77      2.32 0.556
&lt;/code&gt;&lt;/pre&gt;&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-10-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;
&lt;p&gt;Wow! log binomial and modified poisson still win! Look at logistic regression, the exponentiated coefficient is no longer close to RR. So far this still doesn&amp;rsquo;t break log bin, let&amp;rsquo;s push further.&lt;/p&gt;




&lt;h3 id=&#34;very-high-rr&#34;&gt;Very High RR
  &lt;a href=&#34;#very-high-rr&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;p&gt;Let&amp;rsquo;s set RR to 15? base rate of 6%&lt;/p&gt;
&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## y
##   0   1 
## 697 603
&lt;/code&gt;&lt;/pre&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 4 × 5
##   model            estimate conf.low conf.high  width
##   &amp;lt;fct&amp;gt;               &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;  &amp;lt;dbl&amp;gt;
## 1 Poisson              13.3     9.98      18.2   8.26
## 2 Log Binomial         NA      NA         NA    NA   
## 3 Logistic            124.     84.2      187.  103.  
## 4 Modified Poisson     13.3    10.1       17.6   7.57
&lt;/code&gt;&lt;/pre&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## Warning: Removed 1 row containing missing values or values outside the scale range
## (`geom_pointrange()`).
&lt;/code&gt;&lt;/pre&gt;&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-11-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;
&lt;p&gt;umm, ok that&amp;rsquo;s a bit hard to see. We notice that log binomial did not converge and logistic regression is estimating odds ratio, not RR lol. Let&amp;rsquo;s remove logisic regression and compare poisson to modified poisson.&lt;/p&gt;
&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-12-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;
&lt;p&gt;Alright! We&amp;rsquo;ve seen in the past that log binomial and modified poisson estimates and SE are pretty close, and since our log binomial model could not converge, modified Poisson is our winner! 🥇&lt;/p&gt;




&lt;h2 id=&#34;repeated&#34;&gt;Repeat Measures
  &lt;a href=&#34;#repeated&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;p&gt;Alright, now that we&amp;rsquo;re convinced (though we really need to do lots of replicates of the above to &amp;ldquo;convince&amp;rdquo; ourselves that it&amp;rsquo;s not just one dataset that did this - let&amp;rsquo;s spawn an agent to push these to computer cluster and replicate for us while we go on&amp;hellip; 
&lt;a href=&#34;#replicate-of-sim&#34;&gt;if you&amp;rsquo;re curious click here&lt;/a&gt;), let&amp;rsquo;s insert some interesting thing here. What if, we have repeated measure? Let’s simulate 65 subjects with 20 visits each, treatment fixed per subject and a subject-level random effect, then checked how often each method’s 95% CI actually contained the true RR (~0.63).&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;library&lt;/span&gt;(geepack)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;set.seed&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;n_id &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#099&#34;&gt;65&lt;/span&gt;     &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# subjects&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;n_visit &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#099&#34;&gt;20&lt;/span&gt;  &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# visits per subject (65 x 20 = 1300 rows)&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;baseline &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;log&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;0.15&lt;/span&gt;) &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# 15% base rate&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;rr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;log&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;0.6&lt;/span&gt;) &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# RR of 0.6&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;id &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rep&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;&lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;:&lt;/span&gt;n_id, each &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; n_visit)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rep&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rbinom&lt;/span&gt;(n_id, &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;, &lt;span style=&#34;color:#099&#34;&gt;0.5&lt;/span&gt;), each &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; n_visit) &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# treatment fixed per subject&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;u &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rep&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rnorm&lt;/span&gt;(n_id, &lt;span style=&#34;color:#099&#34;&gt;0&lt;/span&gt;, &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;), each &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; n_visit)    &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# subject random effect, strong clustering&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;p &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;pmin&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;exp&lt;/span&gt;(baseline &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt; rr&lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;*&lt;/span&gt;x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt; u), &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rbinom&lt;/span&gt;(n_id &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;*&lt;/span&gt; n_visit, &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;, p)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;df &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tibble&lt;/span&gt;(id, x, y)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;table&lt;/span&gt;(df&lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;$&lt;/span&gt;y)
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## 
##    0    1 
## 1078  222
&lt;/code&gt;&lt;/pre&gt;&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# true marginal RR (pmin caps a few risks at 1, so it&amp;#39;s slightly off 0.6)&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;u_big &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;rnorm&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;1e6&lt;/span&gt;, &lt;span style=&#34;color:#099&#34;&gt;0&lt;/span&gt;, &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;true_rr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mean&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;pmin&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;exp&lt;/span&gt;(baseline &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt; rr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt; u_big), &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;)) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;/&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mean&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;pmin&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;exp&lt;/span&gt;(baseline &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt; u_big), &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;true_rr
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## [1] 0.6271079
&lt;/code&gt;&lt;/pre&gt;&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# models&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;model_poisson_rm &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;glm&lt;/span&gt;(y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; x, family &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; poisson, data &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; df)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;model_gee_rm &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;geeglm&lt;/span&gt;(y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; x, family &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;poisson&lt;/span&gt;(&lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;log&amp;#34;&lt;/span&gt;), data &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; df, id &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; id, corstr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;independence&amp;#34;&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;exp_ct &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;function&lt;/span&gt;(ct) {
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  ct &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;    &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;TRUE&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;    &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;across&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;c&lt;/span&gt;(estimate, conf.low, conf.high), exp))
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;}
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;results_rm &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;bind_rows&lt;/span&gt;(
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(model_poisson_rm, exponentiate &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;, conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Poisson&amp;#34;&lt;/span&gt;),
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;coeftest&lt;/span&gt;(model_poisson_rm, vcov &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;vcovHC&lt;/span&gt;(model_poisson_rm, type &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;HC0&amp;#34;&lt;/span&gt;)) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;exp_ct&lt;/span&gt;() &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Modified Poisson&amp;#34;&lt;/span&gt;),
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;coeftest&lt;/span&gt;(model_poisson_rm, vcov &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;vcovCL&lt;/span&gt;(model_poisson_rm, cluster &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt;id, type &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;HC0&amp;#34;&lt;/span&gt;, cadjust &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;FALSE&lt;/span&gt;)) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;exp_ct&lt;/span&gt;() &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Modified Poisson (cluster)&amp;#34;&lt;/span&gt;),
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tidy&lt;/span&gt;(model_gee_rm, exponentiate &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;, conf.int &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#999&#34;&gt;T&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;GEE&amp;#34;&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;filter&lt;/span&gt;(term &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;==&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;x&amp;#34;&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;select&lt;/span&gt;(model, estimate, std.error, conf.low, conf.high) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(width &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; conf.high &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;-&lt;/span&gt; conf.low,
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;         model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;factor&lt;/span&gt;(model, levels &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;c&lt;/span&gt;(&lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Poisson&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Modified Poisson&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Modified Poisson (cluster)&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;GEE&amp;#34;&lt;/span&gt;)))
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;results_rm
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 4 × 6
##   model                      estimate std.error conf.low conf.high width
##   &amp;lt;fct&amp;gt;                         &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt; &amp;lt;dbl&amp;gt;
## 1 Poisson                        1.13     0.134    0.867      1.47 0.603
## 2 Modified Poisson               1.13     0.122    0.888      1.43 0.547
## 3 Modified Poisson (cluster)     1.13     0.243    0.701      1.82 1.12 
## 4 GEE                            1.13     0.243    0.701      1.82 1.12
&lt;/code&gt;&lt;/pre&gt;&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;results_rm &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;ggplot&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;aes&lt;/span&gt;(x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; model, y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; estimate, ymin &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; conf.low, ymax &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; conf.high)) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;geom_pointrange&lt;/span&gt;() &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;geom_hline&lt;/span&gt;(yintercept &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; true_rr, linetype &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;dashed&amp;#34;&lt;/span&gt;, color &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;red&amp;#34;&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;ylim&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;0&lt;/span&gt;, &lt;span style=&#34;color:#099&#34;&gt;2.5&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-13-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;
&lt;p&gt;Wow! Remember we said poisson SE is bigger than modified poisson, but now both poisson and modified poisson have narrower 95%CI than the clustering methods. Which is right? I mean, we did simulate repeated measures with strong clustering, so the clustering methods should be more appropriate, right? But the 95%CI didn&amp;rsquo;t even cover the truth&amp;hellip; let&amp;rsquo;s call the agent again.&lt;/p&gt;
&lt;p&gt;&lt;img src=&#34;replicates_repeated_measures.png&#34; alt=&#34;&#34;&gt;&lt;/p&gt;
&lt;p&gt;Holy Mac! Yea poisson and modified poisson without clusterings have narrow CI but look at their coverage! Poor! And the clustering methods have wider CI but their coverage is much better, close to 95%. So, we can see that the clustering methods are more appropriate for repeated measures data, even though they have wider confidence intervals. Alright, right now I&amp;rsquo;m pretty convinced about modified poisson and use clustering if appropriate. Next, let&amp;rsquo;s see if we can estimate RR with g-computation and bootstrap by subject to see if we can get a similar result as modified poisson with robust SE + clustering.&lt;/p&gt;




&lt;h3 id=&#34;gcomp&#34;&gt;RR Without Modified Poisson with LR G-computation
  &lt;a href=&#34;#gcomp&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h3&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# g-computation: fit logistic, predict everyone as treated vs untreated, take ratio of mean risks&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;gcomp_rr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;function&lt;/span&gt;(data) {
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  m &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;glm&lt;/span&gt;(y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; x, family &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;binomial&lt;/span&gt;(&lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;logit&amp;#34;&lt;/span&gt;), data &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; data)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  p1 &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mean&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;predict&lt;/span&gt;(m, newdata &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(data, x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;), type &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;response&amp;#34;&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  p0 &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mean&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;predict&lt;/span&gt;(m, newdata &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(data, x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#099&#34;&gt;0&lt;/span&gt;), type &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;response&amp;#34;&lt;/span&gt;))
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  p1 &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;/&lt;/span&gt; p0
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;}
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;rr_gcomp &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;gcomp_rr&lt;/span&gt;(df)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# bootstrap by subject: resample subjects (not rows), keep all their visits together&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;df_split &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;split&lt;/span&gt;(df, df&lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;$&lt;/span&gt;id)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;set.seed&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;boot_rr &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;map_dbl&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;1&lt;/span&gt;&lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;:&lt;/span&gt;&lt;span style=&#34;color:#099&#34;&gt;1000&lt;/span&gt;, &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;~&lt;/span&gt; {
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  samp &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;sample&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;names&lt;/span&gt;(df_split), replace &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;TRUE&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  boot_df &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;bind_rows&lt;/span&gt;(df_split[samp], .id &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;boot_id&amp;#34;&lt;/span&gt;) &lt;span style=&#34;color:#998;font-style:italic&#34;&gt;# new id so duplicated subjects count separately&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;gcomp_rr&lt;/span&gt;(boot_df)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;})
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;results_gcomp &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;&amp;lt;-&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;tibble&lt;/span&gt;(
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;G-comp (cluster bootstrap)&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  estimate &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; rr_gcomp,
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  conf.low &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;quantile&lt;/span&gt;(boot_rr, &lt;span style=&#34;color:#099&#34;&gt;0.025&lt;/span&gt;),
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  conf.high &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;quantile&lt;/span&gt;(boot_rr, &lt;span style=&#34;color:#099&#34;&gt;0.975&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(width &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; conf.high &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;-&lt;/span&gt; conf.low)
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;results_gcomp
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;## # A tibble: 1 × 5
##   model                      estimate conf.low conf.high width
##   &amp;lt;chr&amp;gt;                         &amp;lt;dbl&amp;gt;    &amp;lt;dbl&amp;gt;     &amp;lt;dbl&amp;gt; &amp;lt;dbl&amp;gt;
## 1 G-comp (cluster bootstrap)     1.13    0.690      1.80  1.11
&lt;/code&gt;&lt;/pre&gt;&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;background-color:#fff;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-r&#34; data-lang=&#34;r&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;bind_rows&lt;/span&gt;(results_rm, results_gcomp) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;mutate&lt;/span&gt;(model &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;factor&lt;/span&gt;(model, levels &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;c&lt;/span&gt;(&lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Poisson&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Modified Poisson&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;Modified Poisson (cluster)&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;GEE&amp;#34;&lt;/span&gt;, &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;G-comp (cluster bootstrap)&amp;#34;&lt;/span&gt;))) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;|&amp;gt;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;ggplot&lt;/span&gt;(&lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;aes&lt;/span&gt;(x &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; model, y &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; estimate, ymin &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; conf.low, ymax &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; conf.high)) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;geom_pointrange&lt;/span&gt;() &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;geom_hline&lt;/span&gt;(yintercept &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; true_rr, linetype &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;dashed&amp;#34;&lt;/span&gt;, color &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;=&lt;/span&gt; &lt;span style=&#34;color:#d14&#34;&gt;&amp;#34;red&amp;#34;&lt;/span&gt;) &lt;span style=&#34;color:#000;font-weight:bold&#34;&gt;+&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  &lt;span style=&#34;color:#900;font-weight:bold&#34;&gt;ylim&lt;/span&gt;(&lt;span style=&#34;color:#099&#34;&gt;0&lt;/span&gt;, &lt;span style=&#34;color:#099&#34;&gt;2.5&lt;/span&gt;)
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;img src=&#34;https://www.kenkoonwong.com/blog/modified-poisson/index_files/figure-html/unnamed-chunk-14-1.png&#34; alt=&#34;&#34; width=&#34;672&#34; /&gt;
&lt;p&gt;Look at that! G-computation with logistic regression gives us the exact same RR as modified Poisson. That makes sense, because with only x in the model, both are just computing the risk in the treated divided by the risk in the untreated. And bootstrapping by subject (not by row!) gives a 95% CI almost identical to the cluster robust SE and GEE. Should we call the agent again?&lt;/p&gt;
&lt;p&gt;&lt;img src=&#34;replicates_repeated_measures_gcomp.png&#34; alt=&#34;&#34;&gt;
Et viola! Same same! woo hoo! 🙌&lt;/p&gt;




&lt;h2 id=&#34;replicate-of-sim&#34;&gt;Replicates of Our 4 Scenarios
  &lt;a href=&#34;#replicate-of-sim&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;p&gt;&lt;img src=&#34;replicates_normal.png&#34; alt=&#34;&#34;&gt;
&lt;img src=&#34;replicates_high_event_rate.png&#34; alt=&#34;&#34;&gt;
&lt;img src=&#34;replicates_high_rr.png&#34; alt=&#34;&#34;&gt;
&lt;img src=&#34;replicates_very_high_rr.png&#34; alt=&#34;&#34;&gt;
Now we&amp;rsquo;ve seen it! All 4 scenarios with replicates showed pretty convincing evidence that modified poisson is a good choice for estimating risk ratios, especially when log-binomial regression fails to converge. Not only we care about convergence and narrow CI, we care about appropriate CI with appropriate coverage as we saw from the repeated measures simulation.&lt;/p&gt;




&lt;h2 id=&#34;opportunities&#34;&gt;Opportunities For Improvement
  &lt;a href=&#34;#opportunities&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;No confounders yet! We only had one binary treatment. I heard this is where log-binomial really struggles to converge.&lt;/li&gt;
&lt;li&gt;Explore non-collapsibility of OR&lt;/li&gt;
&lt;li&gt;Other GEE correlations. We only used independence, to match modified Poisson. What about exchangeable?&lt;/li&gt;
&lt;li&gt;Rescuing log-binomial. We let it fail. Starting values might fix it. Is it the model, or just the algorithm?&lt;/li&gt;
&lt;li&gt;Risk difference. We only looked at RR. Absolute risk difference often matters more to patients and clinicians.&lt;/li&gt;
&lt;li&gt;Need to try to solving robust/sandwich SE from scratch to know how it actually works&lt;/li&gt;
&lt;/ul&gt;




&lt;h2 id=&#34;lessons&#34;&gt;Lessons learnt
  &lt;a href=&#34;#lessons&#34;&gt;&lt;svg class=&#34;anchor-symbol&#34; aria-hidden=&#34;true&#34; height=&#34;26&#34; width=&#34;26&#34; viewBox=&#34;0 0 22 22&#34; xmlns=&#34;http://www.w3.org/2000/svg&#34;&gt;
      &lt;path d=&#34;M0 0h24v24H0z&#34; fill=&#34;currentColor&#34;&gt;&lt;/path&gt;
      &lt;path d=&#34;M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z&#34;&gt;&lt;/path&gt;
    &lt;/svg&gt;&lt;/a&gt;
&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;learnt what modified poisson with robust SE is&lt;/li&gt;
&lt;li&gt;learnt about clustering with &lt;code&gt;vcovCl&lt;/code&gt; or &lt;code&gt;geepack&lt;/code&gt; package&lt;/li&gt;
&lt;li&gt;learnt that modified poisson with robuse SE w/wo clustering (depending on data) really matters because SE could be wrong either too narrow or too wide!&lt;/li&gt;
&lt;li&gt;thank goodness to Claude code where I can paste my rmarkdown of my simulation and models so that it can send replicates to cluster and create the plot for me! Saved me a lot of time trying to code from the ground up! Will definitely be using this more!&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;If you like this article:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;please feel free to send me a 
&lt;a href=&#34;https://www.kenkoonwong.com/blog/&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;comment or visit my other blogs&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;please feel free to follow me on 
&lt;a href=&#34;https://bsky.app/profile/kenkoonwong.bsky.social&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;BlueSky&lt;/a&gt;, 
&lt;a href=&#34;https://twitter.com/kenkoonwong/&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;twitter&lt;/a&gt;, 
&lt;a href=&#34;https://github.com/kenkoonwong/&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;GitHub&lt;/a&gt; or 
&lt;a href=&#34;https://rstats.me/@kenkoonwong&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;Mastodon&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;if you would like collaborate please feel free to 
&lt;a href=&#34;https://www.kenkoonwong.com/contact/&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;contact me&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
</description>
    </item>
    
  </channel>
</rss>