Disqus - Latest Comments for andrewheiss

Re: A guide to working with country-year panel data and Bayesian multilevel models | Andrew Heiss

andrewheiss — Tue, 29 Nov 2022 00:25:13 -0000

For (1), check out chapters 15–17 of Bayes Rules (https://www.bayesrulesbook...., which has the best overview of multilevel models I've found (even if you don't use Bayesian methods). They cover the b_0 offset notation (see 16.3.2: https://www.bayesrulesbook.... as well as an alternative type of notation for random effects

For (2), multilevel models in general (both frequentist and Bayesian) don't really use robust se (see https://statmodeling.stat.c.... Rather than account for the clustered structure of the data through the standard errors, the multilevel approach models the clustering directly

Re: A guide to working with country-year panel data and Bayesian multilevel models | Andrew Heiss

andrewheiss — Thu, 15 Sep 2022 15:55:55 -0000

Oh neat! I hadn't seen the performance package yet—that looks really helpful!

Re: Marginalia: A guide to figuring out what the heck marginal effects, marginal slopes, average marginal effects, marginal effects at the mean, and all these other marginal things are | Andrew Heiss

andrewheiss — Thu, 15 Sep 2022 15:53:51 -0000

I would agree :). The AME and MEM are different estimands, which is why they're so different in the results. In real life you probably shouldn't calculate both and compare them to each other

Re: Quick and easy ways to deal with long labels in ggplot2 | Andrew Heiss

andrewheiss — Mon, 12 Sep 2022 13:49:41 -0000

It seems that `label_wrap()` and `str_wrap()` aren't compatible with `expression()`. However, you can fake it if you use ggtext since it can parse a subset of HTML tags like <sub> and <sup>!

Here's a reprex: https://gist.github.com/and...

https://uploads.disquscdn.c...

Re: Convert logistic regression standard errors to odds ratios with R | Andrew Heiss

andrewheiss — Sun, 15 Nov 2020 21:53:13 -0000

broom::tidy() exponentiates the coefficients and confidence intervals, but not the standard errors

https://uploads.disquscdn.c...

Re: Meld regression output from multiple imputations with tidyverse | Andrew Heiss

andrewheiss — Wed, 30 Sep 2020 01:05:50 -0000

Stargazer can't handle them, but there are more modern regression table packages that can handle any model that broom can handle, like modelsummary (https://vincentarelbundock.... or huxtable (https://hughjonesd.github.i...

Re: Ways to close backdoors in DAGs | Andrew Heiss

andrewheiss — Wed, 26 Feb 2020 23:14:46 -0000

That I'm not entirely sure of. I've seen people talk about using any methods to get the most accurate propensity scores (even things like random forests and neural networks), though based on this tweet, those techniques can overfit propensity score models. So maybe include squared and cubed terms? And maybe not? :)

Re: Ways to close backdoors in DAGs | Andrew Heiss

andrewheiss — Tue, 25 Feb 2020 22:20:36 -0000

In this case, just controlling for the backdoors in a regression model like that works just fine. That will only work, however, if all the relationships between the confounders and treatment and outcome are linear, which they are here because I made them that way. If there are any non-linearities or anything that's better modeled non-parametrically, using controls in regular OLS won't work (but IPW and matching should work)

Re: Meld regression output from multiple imputations with tidyverse | Andrew Heiss

andrewheiss — Thu, 05 Sep 2019 11:11:41 -0000

Hi! When summarizing and describing the data, you don't choose a single imputation, but you have to pool them together somehow. There's not really a standard for this—you might report the median, or the mean + standard deviation, or include density plots, etc.

Re: Working with R, Cairo graphics, custom fonts, and ggplot | Andrew Heiss

andrewheiss — Mon, 01 Jul 2019 22:02:46 -0000

I'm not sure—it works for me on 3.6.0 in macOS… Can you post a reproducible example?

Re: How to test any hypothesis with the `infer` package | Andrew Heiss

andrewheiss — Fri, 24 May 2019 15:16:07 -0000

No problem! Also, getting these codes is way easier than reading the CSV if you use the WDI package instead. It connects with the World Bank's API directly and returns pre-tidied data.

Re: How to test any hypothesis with the `infer` package | Andrew Heiss

andrewheiss — Fri, 24 May 2019 15:10:48 -0000

Those are the columns in WWBICountry.csv. The columns in WWBIData.csv should look like this:

[1] "Country Name" "Country Code" "Indicator Name" "Indicator Code" "2000" "2001" [7] "2002" "2003" "2004" "2005" "2006" "2007" [13] "2008" "2009" "2010" "2011" "2012" "2013" [19] "2014" "2015" "2016" "X22"

Re: How to test any hypothesis with the `infer` package | Andrew Heiss

andrewheiss — Fri, 24 May 2019 11:53:34 -0000

It still works. If you download the CSV file from https://datacatalog.worldba..., you'll get two files, one named "WWBICountry.csv" and one named "WWBIData.csv". WWBIData contains all the required data. That file itself doesn't have many of the columns in the code, but that's because the code that makes `wwbi_small` reshapes the data.

Re: Create a cheap, disposable supercomputer with R, DigitalOcean, and future | Andrew Heiss

andrewheiss — Thu, 02 May 2019 14:14:58 -0000

Yeah, the data has to be present on each of the machines

Re: Create a cheap, disposable supercomputer with R, DigitalOcean, and future | Andrew Heiss

andrewheiss — Tue, 30 Apr 2019 11:24:08 -0000

You need to adjust the permissions for ~/.ssh: https://www.howtogeek.com/1...

Re: Tidy text, parts of speech, and unique words in the Bible | Andrew Heiss

andrewheiss — Fri, 22 Feb 2019 12:46:04 -0000

The ESV is still under copyright and not in the public domain, but
there's an API (with a bunch of rate limits) here: https://api.esv.org/

Re: Tidy text, parts of speech, and unique words in the Qur'an | Andrew Heiss

andrewheiss — Sat, 29 Dec 2018 18:29:25 -0000

That's an issue with RStudio, I think. If you look at the data in a new tab, like with View(fatiha_terms), they should show up fine

Re: Tidy text, parts of speech, and unique words in the Qur'an | Andrew Heiss

andrewheiss — Sat, 29 Dec 2018 15:46:17 -0000

Huh. Try running `system.file("extdata", package = "cleanNLP")` to
see what the full path is on your computer, and then open that in
terminal or Finder/File Explorer. You should have a folder named
"stanford-corenlp-full-2018-10-05" and in there, there should be a large
60+ MB file named "stanford-arabic-corenlp-2018-10-05-models.jar"

If it's not there, try running `cnlp_download_corenlp(type = "ar")` in R to make it download just the Arabic models

Re: The academic job search finally comes to an end | Andrew Heiss

andrewheiss — Tue, 18 Dec 2018 01:48:46 -0000

Both are acceptable and "syllabi" is winning (see https://books.google.com/ng... ), but I'm sticking with my Anglicized pluralization for as long as possible :)

Re: The academic job search finally comes to an end | Andrew Heiss

andrewheiss — Tue, 18 Dec 2018 01:42:06 -0000

But also syllabuses :) http://languagelog.ldc.upen...

Re: Create supply and demand economics curves with ggplot2 | Andrew Heiss

andrewheiss — Mon, 11 Dec 2017 21:38:41 -0000

%>% comes from the magrittr package (which is part of the tidyverse) and it lets you pipe left-hand side values forward into expressions that appear on the right-hand side (see this chapter in the R for Data Science book: http://r4ds.had.co.nz/pipes.... It also gets loaded when you load dplyr

curve_intersect() is a custom function that's defined earlier in the blog post.

Re: Create supply and demand economics curves with ggplot2 | Andrew Heiss

andrewheiss — Mon, 11 Dec 2017 18:53:51 -0000

I just ran `Hmisc::bezier(c(1, 8, 9), c(1, 5, 9))` with Hmsic 4.0-3 and it's most definitely there…

Re: Working with R, Cairo graphics, custom fonts, and ggplot | Andrew Heiss

andrewheiss — Thu, 23 Nov 2017 02:17:43 -0000

Yep! Those are great themes

Re: Exploring Minard’s 1812 plot with ggplot2 | Andrew Heiss

andrewheiss — Thu, 24 Aug 2017 01:28:37 -0000

Ha, that's my Americanness showing. Fixed now. Thanks!

Re: Convert logistic regression standard errors to odds ratios with R | Andrew Heiss

andrewheiss — Thu, 06 Apr 2017 10:06:45 -0000

Sure, it's messed up and not entirely useful, but it's what Stata does and, correspondingly, it's what appears in lots of published work in economics and political science. When replicating articles that do report the transformed SEs, it's helpful to be able to transform them to ensure the replicated results match the published results.