--- title: "Probability & Odds (Notes)" subtitle: "STAT 155" author: "Your Name" format: html: toc: true toc-depth: 2 embed-resources: true --- ```{r setup} #| include: false knitr::opts_chunk$set( collapse = TRUE, warning = FALSE, message = FALSE, error = TRUE, fig.height = 2.75, fig.width = 4.25, fig.env = 'figure', fig.pos = 'h', fig.align = 'center') # Use a color blind friendly color palette throughout doc library(tidyverse) cb_palette <- c("#0072B2", "#D55E00","black", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#CC79A7") scale_colour_discrete <- function(...) scale_colour_manual(values = cb_palette, ...) scale_fill_discrete <- function(...) scale_fill_manual(values = cb_palette, ...) theme_set(theme_bw()) ``` ## Exercises {-} **Context:** To begin formally learning about probabilities and odds, we’ll be exploring a dataset containing information on 2,500 singleton (i.e. not twins) births in King County, Washington in 2001. Each row contains information from one birth parent, and there are no birth parents included in the dataset more than once. The main research question this study aimed to answer was whether the [First Steps](https://kingcounty.gov/en/dept/dph/health-safety/health-centers-programs-services/maternity-support-wic/maternity-support-services-infant-case-management) program in King County improved birth outcomes for women from socioeconomically disadvantaged backgrounds. We'll attempt to answer this research question using the information available to us as we go! The variables in this dataset we'll look at more closely for each birth parent are: - `age`: age of birth parent at time of birth (years) - `parity`: number of children the birth parent has given birth to before - `bwt`: birthweight of the child (in grams) - `firstep`: indicator for whether the birth parent participated in the “First Steps” pregnancy program - `gestation`: number of weeks at which birth parent gave birth Run the code below to read in the `firststeps` data, and create a few new variables that we'll explore as well. ```{r} library(tidyverse) firststeps <- read_csv("https://mac-stat.github.io/data/firststeps.csv") %>% mutate(firstchild = ifelse(parity == 0, "Yes", "No"), # Is this the first child this parent has had? low_bwt = ifelse(bwt < 2500, "low", "not low"), preterm = ifelse(gestation < 37, "Yes", "No")) # short gestational period ``` ### Exercise 1: Exploring First Steps enrollment and Gestational Age A baby born prior to 37 weeks is considered premature. In figuring out whether we have evidence that the First Steps program is associated with better birth outcomes than those not in the First Steps program, we can look at whether the individuals in the program are more likely to have preterm babies. Below, we make a 2x2 table in R: ```{r} # 2x2 Table: preterm vs. First Steps firststeps %>% count(preterm, firstep) ``` You may be wondering why this is called a 2x2 table, when it looks as though the table has four rows and three columns. The data can be re-arranged (and usually is, in a formal report) as follows... ```{r} firststeps %>% count(preterm, firstep) %>% pivot_wider(names_from = firstep, values_from = n) ``` ... but it's easier to code up the original way! Use the original table to answer the questions below. a. How many birth parents were enrolled in the First Steps program? Which rows did you use to calculate this number? b. What *percentage* of people in the study were enrolled in the First Steps program? **Recall: there were 2500 participants! You can confirm this by adding up the entire third column of the table** c. How many birth parents *who were enrolled in First Steps* had a premature baby? d. What *percentage* of birth parents in First Steps had a premature baby? Think carefully about the numerator and denominator you use to calculate this! e. What *percentage* of birth parents who had a premature baby were enrolled in First Steps? Think carefully about the numerator and denominator you use to calculate this! ### Exercise 2: Formalizing the probabilities Congratulations! If you've made it to this point, you already intuitively know what *marginal* and *conditional* probabilities are. Formally, - a *marginal* probability, denoted $P(A)$ for an event $A$, is the probability that $A$ occurs overall. You calculated the marginal probability that people were enrolled in First Steps in part (b)! In this case, the denominator used to calculate the probability was the total number of people in the study. - a *conditional* probability, denoted $P(A | B)$ for events $A$ and $B$, is the probability that $A$ occurs **given** that event $B$ occurs. You calculated the conditional probability that a premature baby was born **given** that a parent was in First Steps in part (d)! In this case, the denominator used to calculate the probability was the total number of birth parents in the First Steps program. You also calculated a conditional probability in part (e). Using formal probability notation, write the probabilities you calculated in parts (b), (d), and (e) of the previous exercise as > b. P(First Steps) = ___ > d. P(Preterm | First Steps) = ___ > e. P(___ | ___) = __ NOTE: The conditional probabilities calculated in parts (d) and (e) should not be the same! This is because which event you condition on alters the denominator, and the event you're interested in alters the numerator. ### Exercise 3: Using the probabilities To determine if gestational age differed by enrollment in First Steps, we'll want to calculate and compare two conditional probabilities: - P(Preterm | First Steps) = the conditional probability that a baby is born prematurely given that the parent enrolled in First Steps - P(Preterm | not in First Steps) = the conditional probability that a baby is born prematurely given that the parent did *not* enroll in First Steps a. You already calculated P(Preterm | First Steps). Now use the 2x2 table to calculate the other conditional probability: P(Preterm | not in First Steps) = ___ b. A ratio of conditional probabilities, where the conditioning (first) event is the same for both, tells us how many *times* more likely an event is to occur for one group compared to another. Calculate how many times more likely a birth parent enrolled in First Steps is to have a premature baby compared to birth parents not enrolled in First Steps. $$ \frac{P(\text{Preterm} | \text{First Steps})}{P(\text{Preterm} | \text{Not in First Steps})} = $$ c. Write a two-sentence summary, appropriate for a general audience, summarizing your results in terms of a *ratio of probabilities*. Does gestational age appear to differ greatly by First Steps enrollment? What does this imply about the effectiveness of the First Steps program, if anything? ### Exercise 4: Visualizations To go along with your summary, let's make a visualization! There are three basic options for visualization two categorical variables. All are perfectly valid, but some may be more useful to read than others, and display different information. You'll encounter one other fancier option (called a `mosaic` plot) in the next activity. ```{r} # Side-by-side bar chart firststeps %>% ggplot(aes(firstep, fill = preterm)) + geom_bar(position = "dodge") # Stacked bar chart firststeps %>% ggplot(aes(firstep, fill = preterm)) + geom_bar() # Stacked relative frequency bar chart firststeps %>% ggplot(aes(firstep, fill = preterm)) + geom_bar(position = "fill") ``` **Bonus Question:** Which of the above three plots allows you to directly see the conditional probabilities we calculated previously? ### Exercise 5: Exploring First Steps enrollment and Low birthweights Another birth outcome we can consider when comparing those enrolled in the First Steps program to those not enrolled is birth weight. A baby is considered to have *low* birth weight when birth weight is less than 2500 grams. a. Fill in the code below to make a table comparing `low_bwt` to `firsteps`. ```{r} # 2x2 Table: low_bwt vs. First Steps ``` b. Using the table from part (a), calculate the following conditional probabilities: > P(Low birth weight | First Steps) = ___ > P(Normal birth weight | First Steps) = ___ > P(Low birth weight | Not in First Steps) = ___ > P(Normal birth weight | Not in First Steps) = ___ ### Exercise 6: Odds An additional numerical summary that is often useful when working with indicator variables is *odds*. Letting $p$ be the probability that an event occurs, Odds are defined as $$ Odds = \frac{p}{1 - p} $$ Therefore, if we know $p$, we can calculate the odds that an event happens! Similarly, if we know the odds, we can calculate $p$ using $$ p = Odds / (1 + Odds) $$ We can also calculate odds from our 2x2 (or 3x2, 4x2, ...) tables. In colloquial terms, probabilities are "yes"'s over "total"'s, and odds are "yes"'s over "no's". In pseudo-math: $$ p = \frac{Yes}{Total}, \quad Odds = \frac{Yes}{No} $$ We'll see why odds are especially useful when we have binary outcome variables in a regression model in the next activity. For now, note that they're also commonly used in lots of contexts: sports, gambling, case-control studies, etc. a. Using your probability calculations from the previous exercise, calculate the following *odds* > Odds(Low birth weight | First Steps) = ___ > Odds(Normal birth weight | First Steps) = ___ > Odds(Low birth weight | Not in First Steps) = ___ > Odds(Normal birth weight | Not in First Steps) = ___ b. A ratio of odds (called an **odds ratio**, unsurprisingly) tells us how many times *higher* or *greater* the odds are that an event occurs, comparing one group to another. This might sound irritatingly circular. The key here is that while odds ratios *do* allow us to compare binary/indicator outcomes from one group to one another, they *do not* tell us how much *more likely* an event is to occur comparing those same groups. This is distinct from ratios of probabilities! Calculate the ratio of the odds of having a low-birth-weight baby, comparing those in the First Steps program to those not in the First Steps program (i.e., how many times higher/lower is the odds of having a low-birth-weight baby among those in First Steps as compared to those not in First Steps?) c. Write a two-sentence summary, appropriate for a general audience, summarizing your results in terms of an *odds ratio*. Does birth weight appear to differ greatly by First Steps enrollment? What does this imply about the effectiveness of the First Steps program, if anything? d. To go along with your summary, make one of the three visualization options we tried out in Exercise 4. ```{r} # Insert code here... ``` ## Extra Practice {-} ### Exercise 7: Conditional vs. Marginal probabilities Suppose we select a person at random from the entire global population. For each of the following probabilities, which do you think is bigger? Explain your reasoning. a. P(lung cancer) or P(lung cancer | smoker) b. P(likes McDonald's) or P(likes McDonald's | vegetarian) c. P(smart | Mac grad) or P(Mac grad | smart) ### Exercise 8: Probability practice Let's explore whether birthweight of a baby varies by whether or not it was the *first* child that a mother had, *and* whether this relationship differs by First Steps enrollment. We make a table below: ```{r} firststeps %>% count(firstchild, low_bwt, firstep) ``` a. What is the probability that a mother enrolled in First steps who is having their first child, has a baby who is born at a low birthweight? Calculate your answer, and write it using formal probability notation. > P(___ | ___) = ? b. What is the probability that a mother *not* enrolled in First steps who is having their first child, has a baby who is born at a low birthweight? Calculate your answer, and write it using formal probability notation. > P(___ | ___) = ? c. What is the probability that a mother's first child has a low birthweight? Calculate your answer, and write it using formal probability notation. > P(___ | ___) = ? d. How many times *more likely* is a child to be born at a low birthweight, comparing children who are the first born to those not first born? > P(___ | ___) / P(___ | ___) = ? ## Done! - Finalize your notes: (1) Render your notes to an HTML file; (2) Inspect this HTML in your Viewer -- check that your work translated correctly; and (3) Outside RStudio, navigate to your `inclass_activities` subfolder within your `stat155` folder and locate the HTML file -- you can open it again in your browser. - Clean up your RStudio session: End the rendering process by clicking the 'Stop' button in the 'Background Jobs' pane. - Check the solutions in the course website, at the bottom of the corresponding chapter. - Work on homework and/or any extra practice exercises!