Tweeting the Hits?

Someone sent me an email saying that he liked my little essay, “Descriptive statistics aren’t just for losers.” I had no idea what he was talking about, but it sounded like the kind of thing I’d say, so I searched the blog and found this post, which indeed I really like!

I thanked my correspondent for reminding me of this little article I’d forgotten, and he told me he just learned of it via someone’s tweet.

This made me think: Maybe I should have a twitter feed of nothing but old blog entries. I could just go back to 2004 and then go gradually forward, tweeting the items that I judge to remain of interest.

Does this make sense? Or is there a better way to do this? ALternatively, I could do it as a separate blog, but that seems a bit . . . recursive.

21 thoughts on “Tweeting the Hits?

  1. I’m going to preface this by saying I’m not a huge fan of twitter. I think you could just make a regular (say weekly) post on this blog with a few links to earlier articles that you wrote, and maybe include a few sentences of updated commentary or reflections on the articles. Another possibility that’s still within the framework of the blog would be to add an additional area to the blog that has your favorite posts in each category or grouped by further subtopics.

      • I like that. It has the advantages of keeping everything in the same place (blog) and also allow you to reference (possibly many) previous post by writing a new post they can connect and (if needed) update them.

    • Roger:

      I already link to relevant posts by others when I see them! But perhaps I’ll do an occasional feature plugging the different blogs on the blogroll.

  2. Facebook’s timeline feature is just what you need. Scroll down the page and you’ll encounter all the wonderful ideas/comments from way back when.

    • No, no, you’re missing the point! The old entries are already there on the blog, the issue is getting them out for people to see. Nobody’s gonna be scrolling back thousands of posts to see what was happening in 2004.

      • Well timeline is a bit different from that. It makes it very easy to look back at previous years. And does a sort of highlights reel for each year. Maybe a “best-of” type of section here on the blog. Greatest Hits of Andrew Gelman?

  3. I think it’s a great idea. I think I’d recommend this to friends who I think should have been reading all along (plus, it would be a good way for me to catch up on previous posts I haven’t read.

  4. Just add a page (as opposed to a post) to the blog called “Favorite Posts” and simply link to all the posts you want to bring attention to. Put it right next to the Blogroll. That way it will be there no matter when someone stumbles onto your blog. Twitter is a real time medium so if you really think they are timeless posts, Twitter is not a good way to preserve them, because only people who happen to follow you at that moment will pay attention to the tweets.

  5. 1. Add a new “Favourites” category to the blog, to which you can add articles from time to time.

    2. Get the RSS feed address for that category.

    3. Set up your Twitter account.

    4. Use a service like http://twitterfeed.com/ to automatically pick up new additions to the Favourites category and post them to your Twitter account.

    You may want to start from the oldest and post newer ones as you go along, as some automatic posting services like Twitterfeed won’t pick up any new additions that have a datestamp before the newest article (although there may be a setting in Twitterfeed which allows you to override that.

  6. Consider the following analogy:
    “Data mining is to data dredging as enhanced interrogation is to torture.”
    Unlike most analogies, such as
    3 is to 27 as 4 is to 64,or
    p-value is to Fisher as posterior probability is to Bayes,
    agreement or disagreement is thoroughly ambiguous. What other ambiguous analogies lurk in the statistics world?
    Paul Alper

  7. I think a “memory lane” installment, say a couple of times a month or whatever would be interesting, especially if they were then discussed in relation to current positions on things. I know that many Bayesians I speak to have apparently forgotten that just a few short years ago they were insisting on things such as the Likelihood principle. More generally, it would illuminate how attitudes have changed and perhaps make sense of some current-day positions.

    • Mayo:

      You write, “many Bayesians I speak to have apparently forgotten that just a few short years ago they were insisting on things such as the likelihood principle.”

      Maybe so, but not me! If you go to the index of Bayesian Data Analysis (published 1995), you’ll see, “likelihood principle, misplaced appeal to.”

      • Andrew: You don’t advocate performing inference by way of applying Bayes’s Theorem, apparently—but if you did, as other Bayesians do, or claim to, then you’d nominally accept the likelihood principle, since it follows, and it’s interesting to see how some Bayesians want to wrangle out of it these days*. Bottom line: have there been (interesting) changes regarding some of the foundational issues, such that a review of some older discussions would be valuable? (I say yes) Are there current issues that have been well-hashed out in the past that people may be unaware of, such that a review of some older blogposts would be valuable? (Again, I say yes). That was the point of my comment. Sheesh!
        * I heard from a statistician at a recent conference that some Bayesians are so wedded to the LP that they erroneously think it precludes checking assumptions of models (even though it explicitly assumes the correctness of the model), and some even have trouble allowing that different tests of assumptions might be warranted for Binomial vs Negative Binomial model checking.

        • Mayo:

          Yes, I do advocate performing inference by way of applying Bayes theorem! See my book Bayesian Data Analysis for many many examples. But this inference is conditional on the model. The likelihood principle is not so relevant when checking the model and when building the model. The likelihood principle does not apply when checking the fit of model to data—there you need the sampling distribution—nor should it give the statistician an excuse not to model the data collection process, as we discuss in chapter 7.

          The key is your second point above. The binomial vs negative binomial example is in fact in our book, and we make it clear that the sampling distribution, not just the likelihood, is relevant to model checking. As discussed several times on this blog over the years, one of my big problems with the so-called Fisher-exact test is that it corresponds to a sampling distribution (random table with fixed margins) that almost never occurs in real data collection.

  8. John Cook retweets (or re-posts to G+) old material of his. There’s absolutely nothing wrong with it imho. Twitter / blogs /etc are about “new! now! right now!!!!”. If what you wrote 2 years ago is still relevant right now, there’s nothing offensive at all about tweeting it. (as John demonstrates)

    • According to him there is even a feature in HootSuite where you can upload a .csv filled with your “greatest hits” and it will cycle through posting them automatically.

      I’ll add to what I said before, that anybody who likes your stuff won’t mind 1/50th of their screen being taken up by the reminder of an oldie-but-goodie for however long their twitter stream stays permanent.

Comments are closed.