Go to ground.news/pete to see all sides of every story and stay informed on breaking news. Sign up or subscribe through my link before Oct 31, 2023 for 30% off unlimited access.

I'm afraid I don't. Removing the suspicious data points and saying "look it's still significant" isn't that compelling. It assumes the obviously suspicious data points were the only manipulation. The fact that any part of the data set looks (very) dodgy means the whole set should be regarded with suspicion.

Shouldn’t there just be *zero* manipulated data points

But this "explanation" doesn't explain why there was suspicious data in the first place. Why would some students have duplicate id numbers and why would she switch the rows around? Her defense that here were even more duplicate IDs that they didn't find is rather weak. It just proves that there's more manipulation than they caught. At best, she is extremely careless with her data analysis and it's not up to the standard for a professor at a university.

I’m not at all convinced by this “arguments”. The guys at Data Colada never said they had found an exhaustive list of false entries, so I don’t think it’s really that important that removing the points doesn’t get rid of the full effect, the tendency of the change is good enough. It also stands to reason that, given how badly concealed the supposed manipulation was, the reason for having those entries amongst the other ones is that those values were *edited*, not created anew. If the previous values were evidence against the hypothesis, changing them for values that support the hypothesis will have this exact effect (removing the points still keeps the effect significant).

"beyond a reasonable doubt" is only a standard for criminal charges. for academic fraud "really really fishy" is well enough.

The argument that removing the suspect points doesn't change the overall trend of data isn't as convincing as you think because it assumes, firstly, that the obviously suspect data points are the only ones that were manipulated and, secondly, that these points were added. If the professor chose to alter the data points that most conflicted with her desired outcome, then simply removing those points will not show the original outcome of the study.

She hired a PR firm and is trying to abuse the legal system to intimidate whisteblowers. This should not be rewarded by "reserving our judgement and looking at both sides of the debate", it should be strongly condemned.

7:30 "Ha! You missed another examples that shows I was dishonest, checkmate!"

No, I'm not convinced. They pointed out the most obvious ones. The question is why are there 'irregularities' like that in the study at all? Could it be that she just did a better job on hiding the ones?

This is far too generous. Saying "these results would hold if we removed the suspicious data points" only matters if the allegation is that these data points were manufactured out of nothing. I believe Data Colada suggested that the points were moved from one condition to the other, in which case removing them would be dropping all of the points that tended not to support her hypothesis, so this would still be cheating. Additionally, the reasonable doubt standard is used in criminal trials when someone could end up in prison. In a civil trial, which is much more similar to this case, the standard is "a preponderance of evidence" (at least in the US)

Assuming innocence until proven guilty does not mean that one has to interpret any argument in her favor.

She needs to show that the data was not manipulated by providing an innocent explanation for those unusual rows. The argument you are forwarding from her website is that since the flagged rows only increase the effect, rather than are responsible for it entirely, she must be innocent, but that presumes that the other data is not manipulated - which we have reason to doubt now. The presumption of innocence is lost when there are signs of manipulation, and she needs to address those signs. I agree it may feel less clear cut, but Science is built on trust, and without an explanation for altered data in this pattern we cannot continue with that trust in her.

I want to hear an explanation of why these data ordering and duplicate anomalies are there in the first place. A more convincing case against their approach would have been to explain how these got there in the first place, which eliminates the “suspicious” label altogether. And the motive is obvious: an ordinary significance vs. an extremely high significance will make the paper more compelling.

It seems that her explanation is that regardless of data manipulation her results are the same so no harm no foul. Though the issue is manipulating data in the first place. Arguing that removing the allegedly manipulated data still results in her findings doesn't address the issue of manipulating the data.

I disagree with your overall assessment. Her main argument is that data colada didn't identify all of the problematic rows in their original blog post, which could theoretically introduce bias. However, the Harvard investigation had access to the original files before they were tampered with. Data colada did a recent blog post on September 16 that you didn't mention in your video, where they discuss some of the screenshots from the original files that became available in the course of the trial. Apparently, it could be confirmed that all 8 rows mentioned in the original blog post were meddled with, while they missed an additional 3. Furthermore, there were many more alterations that affected individual Excel cells. Francesca attacking the blog post is irrelevant and a distraction. She needs to explain all the discrepancies between early and late versions of the data files for her studies.

I'm not convinced. "If I removed the suspicious data I get same result" is nonsense. p

Wait a minute: DataColada flagged suspicious data, and now she pointed more suspicious data Is this a TikTok where people are registering on camera their own crimes? 🤣

Elaborating that the data pool contains even more dodgy data results and then accuse Data Cokada of "cherry-picking" examples is no argument in her favour. Not at all.

"Why would I lie if the data would have proven I was right" "Why would a pro athlete dope? why would a famous gamer cheat?" Because the pressure of society causes those of mild success to strive for greater renown, wasn't there an entire other story in this same channel where the other professor who faked studies literally admitted to this reasoning?

9:48 dude 0.01 is 50 times larger than 0.002 - that js a truly enormous difference between results. To argue that they’re basically the same tells me she’s still lying and dissembling about the data fakery. That you accepted it honestly makes me question your own training.

I do not think, this is as compelling of the defence. The data itself was never something that can prove anything. But if someone finds enough consistently weird datapoints that also benefit person who is in charge, investigation was definitely required. And I do not expect org like Harvard to let go someone with no good legal backing, so I assume they got something more legally significant than "data looks weird" argument. If not, someone will pay up for sure

Pete, one issue you have overlooked is the value of suspect data BEFORE they have been manipulated. One can reasonably assume that the suspect data worked against the hypothesis before they were manipulated, and that the hypothesis would be disproved if they were counted in its original state. So it is not enough to simply EXCLUDE suspect data and recreate the statistics, you should rather REVERSE their value.

Having ANY duplicate participant IDs, especially if those rows are not identical, completely invalidates all the data. There is absolutely no logical reason for it unless it's a typo (something like two 8s and no 80), but even then it demonstrates a lack of any effort to validate their own data.

I still have questions about how exactly the participant ID's came to be duplicates... I'm a Ph.D. student who runs a department's student pool at an R1, and it's EXTREMELY odd to me that the ID's were duplicated. In our pool, we give each student a unique 5 digit anonymous ID to use for studies. Then, the researcher adds sequential ID's to the data and removes the initial 5 digit number. SO: The researcher would have to use a generate variable command to create the ordered ID's. In Stata, this command would be "gen subject= _n". I would want to know exactly how they managed to have duplicate responses as well... That seems SO suspicious, or at least it seems they don't know how to use online survey software. Basically, I have many questions about the process, not just the overall results...

Pete, I disagree with you on this one. Go back to the plot at 9:02. Now ask yourself: Are the red dots drawn from a significantly different distribution than the blue points? If you accept a 5% significance level, then the answer is clearly yes. If you draw from a distribution defined by the blue dots there is a less than 5% chance to get the red dots. Also, the argument that there is still a significant effect after excluding the red dots is just misleading, and you fell for it. Classic Defense lawyer’s argument. The claim was never that the manipulation only consisted of adding a few data points out of order. The claim is that the entire data set is manipulated, and the proof is that some entries are out of order. It’s like being caught next the empty cookie jar munching on the last one, then claiming the rest of the cookies were already gone when you arrived.

the why would i fake it if it shows im right argument doesnt work with me. She was pretty highly regarded and i feel like she needed the study to be an overwhelming point proven rather than just a well it looks like this worked kinda thing to keep her reputation up

The question is: Is data tampering less worrying because the data tampering only changed the results significatively instead of a lot? Data tampering IS data tampering. How is that up for debate?

The ex-professor's explanation doesn't really explain anything. How did the same participant IDs happen? How did the CalcChain get out of order? If only one part of data is fishy, even if cherry-picked with malicious intent (which is far from being proven beyond reasonable doubt, don't researchers at Data Colada deserve the same benefit of the doubt), does that mean that the rest of the data is trustworthy?

Sounds like 'I faked it and increased the effect size, but it's OK cause p

"Only include data that is not suspicious." is not convincing. How do we know that all the spreadsheets and the data are not compromised? The fact that she acknowledged that there is compromised data means that we can't fully trust the integrity of the data.

“Innocent until proven guilty beyond reasonable doubt” is a proof standard used in court in criminal cases. The court system has different levels of proof standards for different situations, e.g. in a civil lawsuit they can be lower (relevant for Havard vs. Gino), and for getting things like search warrants they are even lower. More importantly, we in science have much higher standards to prove our findings. I cannot submit to Science and tell the editors “you have to prove beyond reasonable doubt that my results are fabricated!”. No, it works the other way around, I have to bring the evidence to support my paper. If the state ever brings criminal charges against her (misuse of public funds or whatever), then the proof standard you mentioned will be applied and probably not met. But as long as this is not a criminal case, you accepted the proof standard she claimed, which was the most favourable for her, without even thinking on whom the burden of proof lies in this case.

So, if the study mildly bears out her hypothesis, she had no motive to fudge the data to make it look as if it bore it out more dramatically?

This is nothing more than a rich narcissist trying to use spin to avoid accountability. If she's so confident, she should have an independent body replicate the study.. unless the study is BS

I watch a lot of videos about video game fraudsters (as surely many of your viewers do), so here's some food for thought: 1) Her arguments and her behaviour are very much like those of cheaters (e.g. "why would I be faking?", playing the victim card or accusing the critics of foul play). 2) Remember what Karl Jobst said: "Players don't cheat to get faster times, they cheat to get times faster". 3) But also, remember what happened to Dream: yes, he cheated, but most probably unwittingly.

Her explanation for why those rows are not suspicious does not hold water - it boils down to "look, this data you do not accuse me of changing supports the hypothesis!" when the issue is the data that potentially did NOT. Her argument is that if you take the "suspicious" values out then there is still a stat sig difference between the groups -- and that is fine and dandy. But that is only a legit argument if the contention is that these rows were ADDED to an existing dataset (i.e., the other data are real and only those rows were added). Then it would absolutely make no sense to add those data. However, if there were in fact ~100 (or however many) responses and the "suspicious" ones were altered, then it is very important to know what the original values of those responses were -- because if those answers went against the hypothesis, then the study might have been not stat significant (and it would not matter at all whether the remaining respondents' answers show a significant difference because that would be a cherry-picked sample). If the results of the overall sample were NOT significant and you were willing to manipulate them, then it would be very simple to switch group assignments of respondents so that those who did not follow the hypothesis suddenly got placed in the other group, or switch a roughly equal number of answers from both study arms in the direction the hypothesis suggests, etc. Showing the results with the "suspicious" rows removed does not address the second possibility (that the actual results from all respondents were not stat sig and that is why the data manipulation was done). And, really, if one were to commit fraud, it would be easier to just alter existing data than to add cases (esp. since records about how many people completed were likely to exist either in terms of compensation notes or similar). Hope that is clear! Also, and my big issue, is the p-value itself. On a sample of 100, with three groups (so ~30 respondents each) it would be so difficult to get a p as low as reported. In fact, Dr. Gino's defense of herself makes me more suspicious there -- because, yes, smaller samples make it more difficult to get significant p-values, but it would be ESPECIALLY challenging to get such tiny p-values with diminished samples -- and the "suspicious" values increased the significance of the results above what just an increase in sample would be expected to do. Also, the duplicate IDs are a HUGE deal since they matched actual to stated behavior based on those (per the method description in the original article), so given that there are two duplicate IDs, we know that some of those values are incorrect regardless of whether the data are faked. That is unforgivably sloppy. So -- not saying this was or was not fraud, but the defense is lousy. One way to "prove" innocence (yes, I know it cannot be proven) would be to see the original paper responses -- I know it was a while back, but any paper copies saved? (Ps: given the sample sizes, changing the answers of just 3 respondents per group would change the proportion by 10% each, so that impressive 37% vs 67% would become a non-significant 47% vs. 57%).

One thing I find bizarre is that as a project I don't understand why undregrads don't try and replicate these massively popular psychology papers and even if the tests are somewhat poorly done and not as standardized across the schools (being done by 20ish yr old psychology students) if ten universities all can't replicate the results that might indicate there is some kind of problem with the original paper and more scrutiny can be placed on more professionally done studies by maybe higher level students.

Instead of reflecting on her mistakes, she will think that she should improve her data falsification skills.

So she's admitting there was a lot of errors, but "it's still significant". So withdraw the original paper and publish a new one without the multiple errors!

Data Colada would need to be given an opportunity for rebuttal - data manipulation of ANY kind has to be a HUGE RED FLAG as to the integrity of the result and consequently the author - for such a short data Table the claim that (10) obvious irregularities are known to exist seems incredulous - HOW can that occur ? - HOW can anyone hold any confidence on ANY of the data in that Table ? - there appears to be a p-value < 0.05 for statistically significant intellectual FRAUD

I fundamentally disagree. Yes, there's a lawsuit and it will be tried against legal standards. However, the argument "well, they found my data was wrong, but not in all the same ways I found it was wrong and that changes the calc" is bullshit. A professional of this caliber and position is required to do better quality control. You may be able to stipulate mens rea is in favor of innocence, but it's still incompetence and should have never made it to publication. That the results still stand is sheer luck, the greater problem is incompetence, lack of quality control. When dealing with scientific analysis, if you have neither of those then your information is worthless and shouldn't be published. She needs to go. Had she seen this, stated, oh, there's some errors and this needs to be retracted for revision, sure. But staying on the premise of even worse QC is a pass undeserved.

What I frankly don't get is why people use one study, no matter how reputable, as final. Even if everyone agrees it is a good study I would think replication would be done just to make sure the sampling (people) used was objective.

The whole spiel of "assume my innocence pleanse" smells of manipulative to me

Uh oh, looks like she threatened to sue Pete.

My friend is a teacher at a technical college and he has expressed his concerns to me about the future because some of his students have displayed an inability to round numbers consistently. He told me that the school doesn't necessarily do much by way of reprimanding cheaters either. We are now the older generation and are concerned that the buildings of the future may be built and designed by morons and dullards because the system cranks them out in exchange for tuition fees masked by certificates and job placement numbers.

Am I the only one that wonders why this study was done in the first place? Are people running out of seriously important subjects to study?

It still doesn’t explain how such glaring ‘errors’ were made on the first place. She is lucky that these errors were not enough to be of statistical significance, but how does a scientist of this level explain away such sloppy work?

For me those papers are forever tainted. Independent replication needs to happen before I would trust the data

Has this study been replicated successfully? If not, then nothing about this argument would convince me. Far too often you find single studies that are cited but never really replicated, because who is to say she didn't curate the data and forgot to remove the more obvious faked data points?

If the accusation was specifically that the "flagged" data points were outright fabricated, then her defense would make sense. However, it seems more likely that the "flagged" data points were edited (rather than fabricated) because the original value cut against the hypothesis. So the fact that the results are still significant if these data points are excluded does not prove there would have been no motivation to edit the data. Naturally, excluding inconvenient data points can give you a similar result to changing them favorably.

Not buying it. No one ever said the manipulation was solely to generate significant findings. It could just as easily be that she just wanted a stronger P value. She created a strawman to knock down.

She *better* protest her innocence - she won't be able to fake results and publish them any more, and have to get a real job. What a tragedy!

I disagree with applying the "beyond a reasonable doubt" standard. This isn't a criminal court. A preponderance standard is reasonable here.

Pete, I really do not agree with you here. 1. The accusation of cherry-picked data is troubling because she herself picks out even more troubling data points. Just because the data colada team picked out a few of them that are most troubling does not mean that this is cherry-picked. They are examples of problems in the dataset as a whole, of which Gino conveniently picks out even more troubling data points as if to say, "Why would a guilty person pick out even more troubling data points?" this is a defense you often see with people who are guilty, you see this with scammers a lot (no I am not trying to say Gino is a scammer, it's where I have a lot of experience as a skeptic society member). You even see this a lot with legal defenses at court. So It would not surprise me if this is done because this was suggested to her by her legal counsel. And those out-of-order data points were still mostly out of order even if you take those two others out. And if they were conveniently placed there for a reason, which is the accusation in the first place, then those other data points that seem in order don't do much as a counterargument. This argument is actually very weak and very much expected, as the accusation by her of cherry-picking is made very easily by her cherry-picking data as well. 2. That the effect still is statistically significant after deleting the troubling data points is not a good argument at all, because firstly, statistically significant does not in any way mean practically significant, it is not an effect size. It means that the probability of the null hypothesis being true is very low, not that the alternative hypothesis is true, as no frequentist statistics can tell you that with any reasonable certainty. Officially, the interpretation of a p-value is "P values are the probability of observing a sample statistic that is at least as extreme as your sample statistic when you assume that the null hypothesis is true." Secondly, the drop in significance after deleting the 8 or 10 (of more than 100) is quite big for deleting so few data points. Which suggests that there are probably even more troubling aspects of data set in a whole. As @theondone has stated so nicely "It also stands to reason that, given how badly concealed the supposed manipulation was, the reason for having those entries amongst the other ones is that those values were *edited*, not created anew. If the previous values were evidence against the hypothesis, changing them for values that support the hypothesis will have this exact effect (removing the points still keeps the effect significant)." So the defense of Gino is not strong at all and not very convincing. And for me, it matters little if she is guilty or not. I only find the data detective elements of these stories fascinating. That professors of high esteem faked data is after so many examples not at all surprising, but neither would it be all that surprising if the allegations against her were not true, if the case was less strong. After this video and reading some of her website, I still am not convinced of her innocence. statisticsbyjim.com/hypothesis-testing/p-values-misinterpreted/ statisticsbyjim.com/hypothesis-testing/interpreting-p-values/

Using p values to justify that their data isnt manipulated is just dumb.. this arbitrary p value threshold should just go.. thats the flaw in her argument..

It’s about trust, it’s about intention, and it was shown through the fake data. Just because it “worked out” doesn’t negate her intention that destroyed trust.

What confuses me is what is considered significant in Psychology. 101 Samples is just not enough. I know it‘s harder to get more data in some fields. What bugs me is the missing understanding how likely false results are with such a low significance…

You are assuming that the only manipulated data is the one that has been uncovered. It could be that other manipulations where much better hidden. Has this study actually been validated in other experiments? Have other researches found the same conclusion?

Beyond reasonable doubt? Why the criminal code guilt threshold? Why not "By the preponderance of the evidence" or "by the clear and convincing evidence" ...because those are the legal thresholds for civil and fraud cases.

It doesn't strike me as very convincing. While some intro course to statistics may claim that a p < 0.05 is "significant", everybody know that it's not a whole lot of significance. A rock-star professor of pop psychology has a lot of incentive to make her results appear not just a bit better than anecdote, but solid. Merely from examining the papers we likely won't know what exactly happened. Maybe outright fraud, maybe just sloppiness that through some incentives always tends to shift results in one direction. Suing not only her former employers, but independent researchers looking into her work, also reeks of despair -- it's almost certainly not winnable, given the extremely high bar of first-amendment rights, and thus presumably brought for reasons that in a sense are dishonest by themselves, perhaps in order to get the authors to retract in order not to spend their savings on lawyers.

Francesca Gino's argument here is a decent defense because it deflects from some of the core issues. She ignores that the data appears to be manipulated. Why were the cells moved? Her claim of not needing to have manipulated the data for the result is predicated on the assumption that only the out-of-place data could be fake. How do we know that the cells in the correct places were not manipulated? A problem with the suspicious cells is the trend. Just from the manipulated cells, the significance decreases. Gino tries to change the suspicious points with her argument. In the case of 7 vs 5, she has a decent point in terms of interpretation. Her argument of 13 as a duplicate is not convincing because the two 13s have different conditions, majors, and responses whereas the 49s are identical. She has found a way to make the argument from Data Colada weaker, but this paper alone is still suspicious and should warrant investigation in the context of other suspicious findings. Gino is intelligent, unlike the guy at FSU, so I expect her to come up with ways to cast doubt on Data Colada and the investigation by Harvard.

Not that convincing. I'd like to see a model on how likely it is that the data points that look dodgy are also towards the extreme ends of the scale. What it looks like to me now is her saying what the people from dataColada saw as suspicious and that also "completely coincidentally" happened to match up with the extreme ends of the scale are not the only dodgy ones, so she really goes on a hunt for data that "should" be flagged as well. This is a bit bizarre.

From her excuses: "The RAs conducting the study simply stacked the paper copies by condition, and then manually entered the data in the order in which the papers were stacked. The sequence of paper entries tended to be in ascending sequence; this makes sense, since it was probably the same sequence in which the participants performed the tasks". No, it doesn't make sense, because then the sequence would tend to be in DESCENDING order, because of LIFO (last in, first out). And why would you give some cards to the participants, if you could just print those numers on the form itself disguised as that fictitious OMB number at the top rigth? If your research assistants can't even count properly from 1 to 100 - how could you trust any data produced by the study/experiment at all?

As not a researcher with a generally curious mind, I am wondering What if those data points are actually not in favour of her hypothesis; Eliminating them may negate any possibility of rejecting the hypothesis significantly.

What bugs me about this is the fact that the underlying scientific finding doesn't seem to have been refuted or corroborated; the personal drama is overshadowing the question for all involved. Honestly, how hard would it be to do a replication study? A third party with no skin in the game should be able to re-run the experiment. If the underlying principle holds, then the results should be similar, and the study ought to stand. If it doesn't, then we either have a statistical fluke (unlikely given the P rating) or we have fraud.

FYI - beyond a reasonable doubt is a standard of proof for criminal cases. Which is quite as high standard. For civil cases the standard is: on balance of probabilities, which is much lower. So, unless Francesca Gino was facing criminal prosecution, we could apply the lower balance of probability standard.

What the enlightened centrism is this? How embarrassing.

Simple solution: Someone does a replication study and publishes the results. If they're similar, then she was sloppy, but not fraudulent. If not, then her study is either an unlikely fluke or a fraud. Incidentally, I find it strange that she made a distinguished career out of studies like this, whether or not they were fraudulent. This isn't my field, and I don't get why this is so significant.

Kinda bonkers how high academics can get away with these, while i get grilled in my masters course because i used the the default setting of smoothing the curves of the plot

Nevertheless, she is 100% guilty of bad data governance and poor leadership. I’m sorry, I expect more professional behavior from academic leaders than messy undocumented spreadsheets.

In science the standard for identity poor job is no " beyond reasonable doubt " it is just to identify pathetic error

I don’t find her arguments convincing at all. First argument about the duplicated 13: Data Colada obviously grouped the conditions first and then sorted the ID, being forgiving about duplicated ID in different groups! Second argument about manipulating disordered records: it’s not about THOSE values, it’s about possible manipulation and the credibility of the whole dataset! Also, some researchers are obsessed with extremely small p values to show super significant. So the reason for why manipulating is just “chasing for the perfection”.

Why doesn’t she uses these claims int he lawsuit instead of her 5 excuses?

Pete did you happen to reproduce Gino’s last analysis you covered? It’d be surprising to me if DataColada didn’t check whether the main result was still significant without the anomalous rows.

She lost her credibility forever! She will be fired from Harvard and will not get any job at any business school!

Hi Pete. Thank you for your interest in quality assurance. Everybody removes data from their data sets to prove their cases since this type of behaviour is endorsed by ISO 5725. Please let's first get rid of ISO 5725 that causes havoc to science, and then we can talk about starting cases. Further, the p-value refers to your own data set of your own lab, which means that it addresses precision and not accuracy. Hence, there's a huge risk that the results cannot be reproduced, and that is the real problem that should be addressed because that's yet another systemic mistake.

I'm confused why trust her calculations of significance and data analysis now. Just do the data analysis yourself and confirm

Ypu could say, no matter how you turn it, you extremely up the significance. Dont forget that not all tampered lines have to have been found out. You can tamper data without moving data too.

The moral of the story is she should have tried harder if she was gonna cheat

I was wondering how the charts would look with all of the manipulated data removed. Thanks for the update Unfortunately that doesn't clear up the situation very cleanly, as people outside will undoubtedly be skeptical about the entire set rather than just the few points that look suspicious. I was going to go "maybe the findings would still be valid if the manipulated data was removed" -- but then I realized that the manipulated data could have belonged anywhere, so just removing them isn't solving anything if the weight was supposed to go elsewhere, such as the opposite side of where they were moved to. So I'm unfortunately going to have to agree with the others that have said that the entire data set should be treated with suspicion if any part of it has been shown to have been manipulated

Pete, you're looking at this the wrong way: it is not the DataColada🍹assertions that need analysis but the original data and conclusions. Saying I didn't need to massage the data to reach this conclusion in no way demonstrates such a 'session' didn't occur. In spite of small 'p's, the plot does not show significant differences between the 3 conditions.

Thanks to the excellent comments over here I learned that evidence that proves her "guilty beyond a reasonable doubt" isn't required in this case since it's not under criminal law. A "preponderance of evidence" is enough in this case as academic fraud is more similar to civil law, as per the US judicial system. However, can someone explain to me the level of importance of "due process" in this case? She claims that both Harvard and DC broke the preset due process: Harvard because they created and held her to a new academic fraud guideline that was created without her knowledge, and DC because they state in their website that they reach out to the professors when they find anomalies in the data and in this case they sent those reports to HBS, her employer, instead. Does this mean that she has an appropriate legal basis for suing those parties? Thanks in advance!

Well, she may not have been lying. But a 5% p-value to believe something is stupid. If you test 20 stupid assumptions, one of them would give a p-value of 5% just by luck.

Excluding flipped data is not enough. If these points were manipulated, they were changed from a different value. We do not know what the original value was so we cannot conclude whether or not a manipulation alterred the conclusion. The argument you state is the most convincing to you is actually the thinnest and was appropriately buried in the appendicies, or better yet excluded. The built-in excel audit trail was sound evidence and was not convincingly refuted.

Her argument is previously with all that suspicious data, it was p

You missed the mark here. The issue is not about including or excluding the fish entries. The issue is what was the original value of those fish entries. One can assume here that if the original value had been included in the analysis, the result would have disprove her hypothesis.

Unconvinced.

Removing the tempered data points from the analysis proves nothing. They were already doctored to not skew the test results, de facto they are already not affecting the result. This "proof" is laughable. Still not convinced? Say I am computing the mean value of rolling a dice and I have a sample of 10k realizations. I then replace from the sample all 5 and 6 values with a uniform distribution of discrete values in 1 to 4. I will then conclude the mean value of trowing a dice is 2.5. You tell me I doctored the sample and tell me which value I have manipulated, if we exclude them from the sample, we still conclude that the mean of throwing a dice is 2.5 instead of 3.5 ....

Glad to see that 'Academia Strikes Back' is probably coming in at least four parts, I can spot some enjoyable entertainment right there

Poison IVY LEAGUE

Stil, relaying on a sample of 101 people is a low scope. In order to get real statistical mean you need at least 150.

She should explain how the fraudulent data got in there, because it taints the rest of it.

I don’t agree with presuming innocence or guilt. I think we should be seeking the truth of what is healthier for us all.

Floyd Landis had all kinds of arguments and a presentation on why he was innocent of doping. Except, he was guilty. Pete, you need to be a researcher, not a lawyer. Stop using legal terminology, use science. If someone wrongly accused you of faking research, what would you do to prove them wrong? I don’t think you’d complain about academic review processes, and I don’t think you’d rework the analysis using a different subset of the data. I think you would: 1. Provide the original dataset 2. Provide affidavits from the study participants 3. Offer to rerun the study (if you were rich, certainly something you could do) 4. Provide the explanation about why the data looks as it does (if you didn’t fake it, there would have to be some other explanation) I’m sure you can think of other things that this professor isn’t doing. She’s clearly using the “throw anything against the wall and see what sticks” approach that guilty people use.

Did she ever explain WHY those records were out of order though? To me, this sounds like "why would i steal £1000 when i'm a millionaire". Also, these exclusions aren't enough, because when you exclude, you could be excluding extreme outliers that are most against your case in their ORIGINAL value. E.g if i flip a coin 10 times, and get 5 heads 5 tails, I can exclude 5 heads and conclude its a double sided tails coin. so P may not have been 0.028 if the original value of the excluded records were kept in. But she is providing a good defence

We are not a court of law trying crimes, we really don’t owe people the benefit of the doubt criminal court does. There’s a higher standard for the state trying to literally imprison you than the public’s perception of a researcher. You aren’t owed complete presumption of innocence in public opinion.

That’s what I don’t love about media - they shape the crowds opinion towards one “truth”. And when the opposing side tells their arguments, the crowd looks at the arguments from the prior media-shaped “truth” point of view. And instead of listening to arguments presumably with zero prior knowledge, we further shape a biased opinion. Look all of these comments. “I don’t agree, here is why … *something sounding logical*” And this *something sounding logical* have not been processed through lenses of both sides, which would be unbiased. Nor has this *smth* been processed with scientific rigour. I don’t take sides here, but urging people that the world often much more complex than what our “logic” tells us and reaching the truth is astoundingly difficult. Be open minded, and always have doubts not only in others opinion, but yours opinion included.

Note: No one "proves" an hypothesis. Basic understanding of science/statistics/ logic.....

If I understand what is being claimed here, this is very, very stupid. Why would you exclude the points that disagree with her version?

Thanks for this!