3 ms·

https://en.wikipedia.org/wiki/Data_dredging https://en.wikipedia.org/wiki/Data_dredging

rcthompson 9y ago
I have to agree. The lower end of the 95% confidence interval on the hazard ratio is 1.01, which means this result is just barely significant at a p-value of 0.05. (The null hypothesis is a hazard ratio of 1.) After any kind of multiple testing correction, this result is not significant.
dotancohen 9y ago
I'm sure that we'll discover many new things to study with proper dredging. In fact under some conditions it is called Unsupervised Machine Learning and viewed in a positive light. However, as stated, these findings could be used to form hypothesis and then to study them. These findings are _not_ conclusions. Of course, Randal Munroe says it better than I could: https://xkcd.com/882/ https://xkcd.com/882/
okket 9y ago
To be fair, the authors are open about the shortcomings: > further research is needed to replicate these findings, determine their significance, and define the underlying mechanism.
- rcthompson 9y ago
  I don't know that the level of significance reported in the paper even justifies further research.
  - dragonwriter 9y ago
    > To be fair, this is more or less "boiler-plate text" added at the end of every research paper There is usually something discussing future research needs, but it's not always oriented toward “replicate and determine significance”. Though something very much like the note on my this paper is common, for good reason, on papers reporting a novel, borderline significant association with no clear mechanism.
- jaclaz 9y ago
  >To be fair, the authors are open about the shortcomings: >> further research is needed to replicate these findings, determine their significance, and define the underlying mechanism. To be fair, this is more or less "boiler-plate text" added at the end of every research paper, sometimes I would like to see a research paper that ends with "... and that's all about it".

I have to agree. The lower end of the 95% confidence interval on the hazard ratio is 1.01, which means this result is just barely significant at a p-value of 0.05. (The null hypothesis is a hazard ratio of 1.) After any kind of multiple testing correction, this result is not significant.

dotancohen 9y ago

I'm sure that we'll discover many new things to study with proper dredging. In fact under some conditions it is called Unsupervised Machine Learning and viewed in a positive light. However, as stated, these findings could be used to form hypothesis and then to study them. These findings are _not_ conclusions. Of course, Randal Munroe says it better than I could: https://xkcd.com/882/ https://xkcd.com/882/

okket 9y ago

To be fair, the authors are open about the shortcomings: > further research is needed to replicate these findings, determine their significance, and define the underlying mechanism.

rcthompson 9y ago
I don't know that the level of significance reported in the paper even justifies further research.
- dragonwriter 9y ago
  > To be fair, this is more or less "boiler-plate text" added at the end of every research paper There is usually something discussing future research needs, but it's not always oriented toward “replicate and determine significance”. Though something very much like the note on my this paper is common, for good reason, on papers reporting a novel, borderline significant association with no clear mechanism.
jaclaz 9y ago
>To be fair, the authors are open about the shortcomings: >> further research is needed to replicate these findings, determine their significance, and define the underlying mechanism. To be fair, this is more or less "boiler-plate text" added at the end of every research paper, sometimes I would like to see a research paper that ends with "... and that's all about it".