Awesome .. 670 responses to my survey of changing practices - forms.gle/kcwQSd8HV7sAhn… ... thank you all so much. I need to get this upto about 1,000 in order to stand a chance of picking out the weak signals, so please share it. Takes about 10-15 minutes. All help appreciated.
Me : Over 800 responses now to the survey, thank you - forms.gle/au9kLWLia3qQJp…
X : Will you publish the data?
Me : No. I'll publish the results, the test of the hypothesis.
X : What if I want to look for something else?
Me : Give me your hypothesis and I'll test it.
X : Will you publish the data?
Me : No. I'll publish the results, the test of the hypothesis.
X : What if I want to look for something else?
Me : Give me your hypothesis and I'll test it.
X : I was more hoping to find interesting correlations.
Me : Oh, hell no. You can always find correlations if you go looking for correlations. Madness lies that way. It's why you build a hypothesis, collect data, test the hypothesis.
X : What if it's wrong.
Me : Then it's wrong.
Me : Oh, hell no. You can always find correlations if you go looking for correlations. Madness lies that way. It's why you build a hypothesis, collect data, test the hypothesis.
X : What if it's wrong.
Me : Then it's wrong.
X : But there could be ...
Me : No. Once the person has run the hypothesis, they are then polluted. You don't go looking for other correlations in a data set. You get one shot. You can run another's hypothesis but then again, it's a one shot deal. I'm religious about this.
Me : No. Once the person has run the hypothesis, they are then polluted. You don't go looking for other correlations in a data set. You get one shot. You can run another's hypothesis but then again, it's a one shot deal. I'm religious about this.
Me : If you want to go looking for correlations in a data set then once you found a correlation, you discard the entire dataset, build your hypothesis on your correlation and collect an entirely new dataset to test it. You never say "I found this correlation in this data".
X : So, I could use the dataset to find a correlation but only if I then go create an entirely new dataset to test it?
Me : Yep and never use the first dataset as any form of justification for your correlation.
X : So, can I?
Me : No. It's a trust issue. I'll run your hypothesis.
Me : Yep and never use the first dataset as any form of justification for your correlation.
X : So, can I?
Me : No. It's a trust issue. I'll run your hypothesis.
X : But I don't know what I'm looking for!
Me : Yep, and that's not filling me with confidence. Do you know how often people go looking for correlations in a dataset and then use the dataset as justification of the correlation?
X : A lot?
Me : In business, all the bloody time.
Me : Yep, and that's not filling me with confidence. Do you know how often people go looking for correlations in a dataset and then use the dataset as justification of the correlation?
X : A lot?
Me : In business, all the bloody time.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
