specification
The particular choices of variables, functional form, and sample in a statistical model; different specifications can yield different results.
As defined in Betsey Stevenson and Justin Wolfers on Happiness, Growth, and the Reinhart-Rogoff Controversy
Where it comes up · 9
- Betsey Stevenson and Justin Wolfers on Happiness, Growth, and the Reinhart-Rogoff Controversy3 claims
- Michael Brendan Dougherty on My Father Left Me Ireland2 claims
- Louis Menand on Psychiatry 05/31/20101 claim
- Sherwin-Williams: Brushstrokes of Success1 claim
- Ed Yong on Science, Replication, and Journalism1 claim
- Jacob Vigdor on the Seattle Minimum Wage1 claim
- Adam Sandow - The Power of Print Media - [Invest Like the Best, EP.376]1 claim
- Lex Fridman Podcast #494 - Jensen Huang1 claim
and 1 more
Most important claims · 10 of 12Search all claims for “specification” →
- Researchers should give away their data in the form they used it, show every decision they made, and provide a running tally of all regressions run — not just a vague 'we tried other specifications that didn't change our results' — so reviewers can see how the dish was made and how many times the researcher went cherry-picking; honest disclosure of how many regressions were run creates real credibility.Betsey Stevenson and Justin Wolfers on Happiness, Growth, and the Reinhart-Rogoff Controversy
- Coding is now fundamentally specification—telling a computer what to build, optionally with an architecture—so the number of people who can code may rise from about 30 million to roughly 1 billion; every carpenter, accountant, plumber and farmer with AI becomes an elevated professional (a carpenter becomes an architect), so the number of programmers will increase rather than decrease.Lex Fridman Podcast #494 - Jensen Huang
- Pre-specifying analyses — declaring in advance which regressions you'll run and which tables you'll present, as in the Oregon Medicaid Experiment before any data came back — is the clearest way to build credibility. It is easier with experiments; for observational data that have existed for decades, building trust that researchers only ran the regressions they said they would is much harder, and self-publishing data and code (the Stiglerian self-regulation approach) is a partial substitute.Betsey Stevenson and Justin Wolfers on Happiness, Growth, and the Reinhart-Rogoff Controversy
- The word 'replication' is widely misunderstood; the broader and more important sense is taking the same question and answering it a different way with different data to see whether results are robust or highly sensitive to the author's chosen data, specifications, and initial conditions — not merely re-running the original author's exact dataset.Betsey Stevenson and Justin Wolfers on Happiness, Growth, and the Reinhart-Rogoff Controversy
- Averaging biased views doesn't always work. As UCLA econometrician Ed Leamer showed, with the same death-penalty data set a proponent can reasonably show each execution deters 30 murders, while an opponent, using a different specification and slightly different variables, can show each execution causes 12 murders. Leamer's conclusion is not to take the average but that the data don't let us answer the question, and Roberts thinks the same may be true of depression treatments.Louis Menand on Psychiatry 05/31/2010
- Published empirical work hides the 'kitchen' process: researchers who run a single specification usually get nothing significant, then convince themselves they had the wrong specification and iteratively add or drop variables and nonlinear terms until they sculpt a publishable result, so readers cannot judge whether the finding is robust.Ed Leamer on the State of Econometrics 05/10/2010
- To avoid cherry-picking results, the researchers committed to writing down their specification before conducting the analysis—a best practice—rather than running a hundred regressions and selecting the preferred outcome, with alternate specifications and excluded industries documented transparently in appendices.Jacob Vigdor on the Seattle Minimum Wage
- Industrial coatings margins depend on end-market customer concentration, the labor-to-material cost mix, and how technical or specified the coating is, which is why some segments are close to commodities while others are highly differentiated.Sherwin-Williams: Brushstrokes of Success
- Physical sampling remains essential in materials purchasing because designers still need to see and touch samples before making large specification decisions, so better digital tools cannot fully replace it.Adam Sandow - The Power of Print Media - [Invest Like the Best, EP.376]
- The 'studies show' discourse in politics is anti-democratic priest-craft: it forbids common people from saying what they want by requiring them to know social science inside-out to opine, while many studies rely on tiny sample sizes and—per Levitt's own admission of running many specifications and reporting only a few—amount to cherry-picked confirmations of predetermined theses, constituting a kind of intellectual fraud that drapes opinions in a scientific gloss.Michael Brendan Dougherty on My Father Left Me Ireland