T-statistic
A test statistic comparing an estimate to its standard error; a value above 2 roughly corresponds to statistical significance at the 0.05 level.
As defined in John Ioannidis on Statistical Significance, Economics, and Replication
Other definitions
A statistic measuring how likely it is that an estimated relationship is due to chance; a high value suggests the relationship is real rather than a fluke.
Ed Leamer on the State of Econometrics 05/10/2010
Where it comes up · 6
- Ed Leamer on the State of Econometrics 05/10/20103 claims
- Machine Learning in Investing2 claims
- Lee Ohanian on the Great Recession and the Labor Market1 claim
- Campbell Harvey on Randomness, Skill, and Investment Strategies1 claim
- John Ioannidis on Statistical Significance, Economics, and Replication1 claim
- Hal Varian on Technology 07/28/20081 claim
Most important claims · 9 of 9Search all claims for “T-statistic” →
- Differences in marginal tax rates that affect the return to work explain a large share of why total hours worked per working-age population in Germany, France, and Northern Europe fell from above U.S. levels in the 1950s-60s to about 20-25% below the U.S. today; in regressions, adding unionization, employment protection, and other labor variables barely changes the tax coefficient's magnitude or t-statistic or the R-squared.Lee Ohanian on the Great Recession and the Labor Market
- The Sharpe Ratio is the strategy's excess return divided by its volatility, and it is mathematically linked to the t-statistic, so a high Sharpe Ratio corresponds to a high t-statistic indicating returns significantly greater than zero.Campbell Harvey on Randomness, Skill, and Investment Strategies
- Statistical significance (a high T-statistic) only indicates that a relationship is measurable and unlikely due to chance; it says nothing about whether the effect is economically important in magnitude, and conflating 'significant' with 'important' is a key error economists make.Ed Leamer on the State of Econometrics 05/10/2010
- Universities, especially at the undergraduate level, struggle to teach things that don't lend themselves to exams, so they over-emphasize testable facts and proofs and miss the subtler, more valuable wisdom—such as understanding that statistical significance is more than a t-statistic being greater than two.Hal Varian on Technology 07/28/2008
- Academic finance models were designed to explore parameters such as whether a beta is significant, while machine learning models are used purely for their output and pragmatic predictive value, so parameter-level interpretation is often sidelined.Machine Learning in Investing
- Even when the final econometric model is silly and the final table of T-statistics is irrelevant, the process of analyzing a complex dataset (like doing theory) helps economists think more clearly and form judgments, and the social conversations help the profession collectively make progress—often not to the right conclusions, but with some scope for progress.Ed Leamer on the State of Econometrics 05/10/2010
- A parameter in an ML model may be useful only in rare contexts, such as emails mentioning Nigeria or princes, and sit dormant the rest of the time, which is hard to capture with standard average-based statistical confidence measures.Machine Learning in Investing
- In econometrics, the goal has traditionally been to find a t-statistic greater than 2, which corresponds to statistical significance at a p-value of 0.05 or smaller, and most published results using multivariate statistical analysis must clear that hurdle to be publishable.John Ioannidis on Statistical Significance, Economics, and Replication
- A T-statistic is a measure of how likely or unlikely it is that a relationship found in data is due to chance, with a high T-statistic meaning the relationship is very likely real rather than a fluke.Ed Leamer on the State of Econometrics 05/10/2010