Leif Nelson

Or you run a study where there's three treatments, condition A, condition B, and condition C, but in the end you drop condition B and you don't even talk about it, you just compare A to C.

Or you run a study where there's three treatments, condition A, condition B, and condition C, but in the end you drop condition B and you don't even talk about it, you just compare A to C.

Or you run a study where there's three treatments, condition A, condition B, and condition C, but in the end you drop condition B and you don't even talk about it, you just compare A to C.

And then there are things that are mildly statistical, but in a very relaxed way. Well, we collected this data, but it's kind of skewed. It has some outliers. and you say, we should eliminate those outliers, or we should Windsorize the outliers, which is basically truncating them down to a lower high number.

And then there are things that are mildly statistical, but in a very relaxed way. Well, we collected this data, but it's kind of skewed. It has some outliers. and you say, we should eliminate those outliers, or we should Windsorize the outliers, which is basically truncating them down to a lower high number.

And then there are things that are mildly statistical, but in a very relaxed way. Well, we collected this data, but it's kind of skewed. It has some outliers. and you say, we should eliminate those outliers, or we should Windsorize the outliers, which is basically truncating them down to a lower high number.

Or you could run them through an algorithm where you say, oh, let's transform them with a logarithm or with a square root. And those are all decisions that are justifiable. They're not crazy. It's just, if you have a consideration of reporting one variable or the other,

Or you could run them through an algorithm where you say, oh, let's transform them with a logarithm or with a square root. And those are all decisions that are justifiable. They're not crazy. It's just, if you have a consideration of reporting one variable or the other,

Or you could run them through an algorithm where you say, oh, let's transform them with a logarithm or with a square root. And those are all decisions that are justifiable. They're not crazy. It's just, if you have a consideration of reporting one variable or the other,

and one variable makes your hypothesis look good, and the other variable makes your hypothesis look less good, you end up reporting the one that looks good, either because you're being self-serving, or honestly, because you'd say, like, I'm not sure which one is better, but my hypothesis tells me it should be the one that looks good, and that one looks good. It's probably the better measure.

and one variable makes your hypothesis look good, and the other variable makes your hypothesis look less good, you end up reporting the one that looks good, either because you're being self-serving, or honestly, because you'd say, like, I'm not sure which one is better, but my hypothesis tells me it should be the one that looks good, and that one looks good. It's probably the better measure.

and one variable makes your hypothesis look good, and the other variable makes your hypothesis look less good, you end up reporting the one that looks good, either because you're being self-serving, or honestly, because you'd say, like, I'm not sure which one is better, but my hypothesis tells me it should be the one that looks good, and that one looks good. It's probably the better measure.

And here's Simonson.

And here's Simonson.

And here's Simonson.

These will be things that can be as simple as a typo. where someone's writing up their report and the means are actually 5.1 and 5.12, but instead someone writes it down as 51.2. And you're like, wow, that's a huge effect, right? And no one corrects it because it's a huge effect in the direction that they were expecting. And so literally a typo might end up in print.

These will be things that can be as simple as a typo. where someone's writing up their report and the means are actually 5.1 and 5.12, but instead someone writes it down as 51.2. And you're like, wow, that's a huge effect, right? And no one corrects it because it's a huge effect in the direction that they were expecting. And so literally a typo might end up in print.

These will be things that can be as simple as a typo. where someone's writing up their report and the means are actually 5.1 and 5.12, but instead someone writes it down as 51.2. And you're like, wow, that's a huge effect, right? And no one corrects it because it's a huge effect in the direction that they were expecting. And so literally a typo might end up in print.

And that's before we get to anything like fraud, like the active fabrication of data or manipulation of data.

And that's before we get to anything like fraud, like the active fabrication of data or manipulation of data.