I took notes in tired/wired format from the *incredible* #rstudioconf workshop "What they forgot to teach you about R" by @JennyBryan and @jimhester_. Here are some highlights, in case useful for others.
— Travis Gerke (@travisgerke) January 18, 2019
stats337/README.md at master · hadley/stats337 · GitHub
— Read on github.com/hadley/stats337/blob/master/README.md
another reason to be careful about what to control for...a post showing how controlling for SES differences on success in an exam based on test prep
not so standard deviation podcast. Many apply but something about confer Kie season and then episode 54 spot for Microsoft https://itunes.apple.com/us/podcast/not-so-standard-deviations/id1040614570?mt=2
Sometimes easy concepts are made so easy you can get confused.
I’m reading a book titled Statistical Rethinking: a bayesian course with examples in R and Stan, written by Richard McElreath.
I really appreciate this book. I feel like the way that I was taught statistics in psychology really missed the mark. I like how the author in this book frames the differences between bayesian in frequentist statisticians, basically reducing them to their goals.
I’m having to read slowly–mostly because we have a newborn in the house and I’m very much sleep derived. But I also want to make sure that I understand what I’m reading and I think the author did a very good job of setting up the purpose of grid approximation, it’s just that for some reason I had a hard time understanding what the they were for.
It’s basically a way to start generating, what’s known in Bayesian inference, a prior.
But even that is mysterious. Let’s say you have some method for making predictions. This method takes several variables to work. If you know the values for these variables you can use your method and make your prediction.
Well, what if you don’t know all the variables? You’ll need to estimate one or several. And the prior is what begins this process.
After you’ve completed a Bayesian analysis, the output will be a distribution of values for that unknown variable. This distribution will give you some ability to quantify your belief in that value.
A Bayesian model is going to mechanically churn out a probability for the phenomenon but before it does that it needs a starting condition, and this is the prior.
The book offers a few ways to obtain a prior, and the author begins with grid_approximation.
In my own words, it’s like taking a prior probability but then spreading it across a range. This range is then fed into the Bayesian model.
In other words, if you think the prior is a 50/50 situation, that each outcome is equally likely, you can spread that assumption across a grid and let the model try on the range of scores.
The book has a thought experiment trying to estimate water coverage over a hypothetical globe. Let’s say the data is 6 w in 9 observations.
This set up fits nicely to a binomial distribution, but what are the parameters? Should Pr(water) =0.5?
Doing this you’d calculate a Bayesian posterior of
But if you took that 0.5 and spread it across a range from 0 to 1, you’d get a distribution of scores, and with a distribution you have more options to study the effectiveness of your model.
Just spreading out that .5 belief into 5 different values from 0 to 1 we get this data.frame, where the first column represents a prior whole the second column is the posterior estimate for the prior:
[1,] 0.00 0.00000000
[2,] 0.25 0.02129338
[3,] 0.50 0.40378549
[4,] 0.75 0.57492114
[5,] 1.00 0.00000000
So, a better estimate.
Spreading the prior across 50 intervals gets you a distribution like this:
The posterior score in the y axis will continue to shrink as the x axis goes to infinity–this is a property of probability distributions. The probability of a function is basically the area under the curve and since a single point can not have area under it, just a line, it’s probability is zero. Of course we can take a range.
Anyway, trying to wrap my head around priors and their estimates. The distribution above, a posterior distribution, is simply a distribution of possible values for a variable in the model. This variable is called a parameter and it is what we are trying to estimate.
I’m in the process of resurrecting a stats course for potential psych majors in community college. I finished Grad school in 1999 for clinical Psychology. Though it was a good program, I find the first year stats sequence uninspiring. Something was missing in the teaching, or maybe the professor was grinding some conceptual axe with those not in the room.
I’d already taken part of a first year sequence, and that professor really brought home the idea that there was still a lot of debate and interpretation on how best to perform a statistical test. Perhaps this was my first memory of political disagreements. So when I started grad school officially, I was turned off by the certainty of everything that the instructor projected.
So, here I am now trying to think about exposing students to potentially their first course in stats. What does this have to do with p-values and bootstrapping?
Well, psych students are often taught a flow chart style of statistical test choosing, and P-values are just the thing that tells you whether you have a significant effect. We tend to skip over the intuitive understanding of what is meant by ‘finding an effect equally or more extreme than the sample statistic assuming the null hypothesis is correct, or true.” Concretely I think students see a goal post of a given alpha level, say .05, and if you pass the goal post, yay!, you found an effect.
But that misses some intuitive appreciation for p-values.
Recently I’ve been lurking in some coursera courses, in particular from the John Hopkins Group of Peng, Leek, and Caffo, and they’ve inspired me to dig a bit more into calculus and probability theory. Along the way, and maybe some return trips to their teachings, I came across some introductions to bootstrapping and how one could calculate an empirical p-value by averaging the number of bootstrapped estimates (of say a median) greater than the observed statistic (again, the sample median). In other words, the average of a Boolean vector (T,F,T,T,T,F…), where you treat the T = 1, F = 0, is just the proportion of True values in the vector.
So, this might be something I incorporate earlier in a stats class then later. Bootstrapping is advanced. But maybe the counting isn’t?
What do you do with data? Do you obtain it? Store it? Analyze it? Interpret? Present?
I sometimes struggle with priorities. Not the big ones, but for instance, should I first learn more python and webscraping, or should I think more about Bayes.
Should I take a course on R and time series, or should I make an SQLdatabase.
Should I try to analyze tweets? To what end? I’m interested in politics, government transparency, gerrymandering, basketball, fighting animal abuse, tracking how money flows into political coffers, so many things.