Thread by @ProfDFrancis

Cardiology registrars or consultants wanting Cardiac CT to level 2 might be interested in this 6-day hands-on small-group course.

Sadly there will be no reasonable punishment for mistakes, such as public humiliation or twittersassination..

eventbrite.co.uk/e/cardiac-ct-l…

Cruelly, Ben Ariff, who is course director, refused my offer of a lecture on "How to make up your data and not get caught."

"There's no need for that sort of thing!", he hastily replies.

I beg to differ.

On an unrelated note, have a look at this for a sizzler of a story!

retractionwatch.com/2020/09/28/maj…

Remember this?

In 2019 people thought I had actually bought a BMW or something.

https://twitter.com/ProfDFrancis/status/1132739796957319169?s=20

But what I meant was this.

You know when you want to get a dataset like (for example) this?

But your actual data look something like this:

Should you:

Our Data Enhancement practitioners at Francis Industries have studied the methods used by a variety of amateurs - and their failings - and have developed an armoury of impregnable methods.

There are two rules of data fiddling.

Rule 1. Do not get caught.
Rule 2. Remember Rule 1.

Francis Industries
Data Enhancement Manual Volume 1 Chapter 1.

"When the panel comes after you, they will use Tried and Tested methods (TM).

Remember, that means the methods have been Tried and Tested -- but not that they work!

Use that predictability to your advantage."

A common first step is that they will do source data verification. That means they will pick on each point and say, "Show me the raw data for this guy here. Show me the Echo [or whatever]. I want to see the rawest possible data."

Good thing to ask for?

Well, it catches out amateurs who make foolish mistakes.

When they come and ask for the raw data, SHOULD YOU GIVE IT TO THEM?

I mean, it's so obvious.

You have lots of options, even if you have made up the data.

For a small fee, Francis Industries offers Plan A.

In short, this is that a flood has washed away all the raw data.

Our full package (only $9,995 per paper) provides fuller details.

For example, this is particularly efficacious in low-altitude countries, such as the Netherlands.

NEVER hand over the raw data for a study you have made up. That is just daft.

(We give you this tip for free, cos we like you.)

Because the panel are expecting to leaf through data, if you don't give them any data, they can't make any conclusions.

That's the hilarious thing.

Even if you have claimed to have cured cancer - when they ask, you can just shrug, and can only be "done" for not keeping records.

If they somehow get hold of the raw data, and realise it doesn't match up, then you are in real doodoo.

Time for Plan B.

All arrangements will be made for you by Francis Emergency Travel Agencies

What happens next?

So now you are settled in a new environment, safe from extradition - since making up stuff is not taken seriously enough for it.

It's not over.

They can do a video interview!

Francis Industries can guide you through this process.

In their December 2012 report on the DECREASE fun-and-games, the Erasmus university committee left readers in no doubt that they bent over backwards to get useful evidence.

Of course, this is all a lot of trouble for young researchers.

Having to fake climate disasters or get a Green Card, is a time consuming and irritating distraction, from more amusing and academically productive activities like perfecting your Gel-shopping technique.

For only $99,995, we offer the Silver Level plan for faking data without getting caught during an investigation.

The key is that they only look for source data points that you show.

They DO NOT check that there aren't data points you haven't shown.

So the Silver Level plan, which works in any specialty, is to simply delete the points that mess up your correlation.

Tada!

So when the panel come to you and point out a data point, like this one, and say, "OK let's have the source for this!"

You can just point them to your data archive, and tell them where to look.

At no stage will they point to an empty space and say,

"Where was the data point you removed from here?"

That just doesn't happen.

Can you imagine a university teachder going up to a blank blackboard and demanding you answer, "why is there nothing here?"

Well, the panel is composed of teachers, not crooks.

They look at things that are there, not at things that have been removed!

Francis Industries does not hire scientists for its Data Augmentation division. They have the wrong training.

It recruits its staff elsewhere.

Our staff don't even carry car keys. This is how they get into (even their own) cars.

Faced with a plot like this,

They would not take the traditional panelists approach of picking a datapoint and asking for raw data.

They wouldn't do this:

Instead they would comment on this.

Comment 1.

Comment 2.

"Nice try here and here, to leave some distractors, to hide the sharp cutoff, but not really professional standard."

We don't recommend the "sharp" cutoff style any more.

Tends to draw attention.

We recommend a more subtle approach.

Just "thin out" the unwanted segments a little. Not nothing, just less.

Otherwise we will recognise your handiwork, like a car thief recognises a car window that has already been jimmied, doesn't waste time breaking into a vehicle that has already been looted.

OK enough free education!

More one weekend, when we go through the Gold Level plan: no deletion of patients, and therefore evades even super-rigorous investigation panels.

Yes. It may lead to long rhubarb in up to 70% of people.

On an unrelated note, Francis Industries can also advise on what statements to make regarding any allegations.

For example, we do not recommend this approach.

And never, after being caught out on what people can check, then expect people to believe you on things they CAN'T.

No-one is going to take your conclusion-forming skill, or even your raw data acquisition, seriously at this stage.

And if you have been caught in a FrankenFigure, made of various bits of different patient but presented as though they are all the same patient,

Never say that the FrankenFigure could have been assembled lots of different ways, and that this was just the most artistically pleasing.

The Rubik's-cube-like nature of FrankenFigures, with billions of combinations possible to display, is nothing to be proud of.

Don't bank on the panelists to be so elderly that they have never used a computer.

Having different image programs for different images doesn't change John Smith into Vijay Patel.

Full explanation given here.

(This approach is not recommended by Francis Industries)

retractionwatch.com/wp-content/upl…

If you are a straightforward, honest researcher, when you realise there are substantial misrepresentations in your dataset, what should you do?

If you choose Path 1, the ONLY criticism anyone can make of you is that you made a mistake.

Pretty puny criticism. The only people who don't make mistakes are people who don't do anything at all.

The second best option is "Ignore it quietly"

For many papers, nobody will notice.

For example, can anyone see the mistake in this paper?

scirp.org/pdf/AM20111200…

In general, if ANY of your co-authors wants to retract the paper, what should you do?

When, of an N-author paper, N-1 authors are all agreed that it should be retracted, what should you do?

NEVER be that lone person brazening it out, when all sensible people have voted with their feet.

So I leave you with this parting thought.

Look through all your field's papers.
Find the "diagnostic accuracy" plots that look like this.

Francis Motoren Werke GmbH, my wholly owned EU-based Data Enhancement subsidiary, uses the following trademark, to remind people to AVOID this instantly recognisable gaffe.

Annoyingly a car manufacturer liked it so much they stole it (with reversed colors). We need a lawyer!

Oh wow! A fellow old-timer!

Yes, I was taken in by this too, all those years ago.

Who could ever forget:

"The plot that sank a thousand PhDs"

And poor old Wilfried Mullens!

He was working at Cleveland Clinic and thought he might just do a little study to check if it was true.

Was it true?

Source link for Wilfried Mullens paper:

ahajournals.org/doi/pdf/10.116…

So what did the great and the good of the community say?

Well, let's have a look, shall we?

How would you rate correspondent #1's remark?

What does it say?

Correspondent #2 is brilliant!

I am going to make a note of this and use the ploy myself

1. You fail to correlate because you are stupid in making your measurement

2. Anyway if it doesn't correlate it is because the old measure is wrong and the new one better

3. Stop thinking about the old measurement

WIN!

Correspondent #3 makes a more intelligent effort at stabbing Wilf.

The argument is:

1. Some people say do it one way, others another way. You did it both and took the average, to reduce the noise.
We don't like that it didn't work when you did that.

2. Also we don't like your cutoff.(Even though non- correlation is nothing to do with a cutoff)

The final correspondent is the best, though, highlighting the method that Francis Industries will be proud of.

If a value is higher than you expect from previous work

And if a value is lower than you expect from previous work

The argument presented is clever. The correspondent is saying that when the spectrum is broad (doppler image fuzzy) you should read off the value that is most consistent with your expectations.

People do that in clinical practice, but in research this is called cheating.

Anyway, following this discovery that the moon wasn't indeed made of green cheese, nor is E/E' the dominant determinant of PCWP, what happened to Wilfried Mullens?

He's very happy there now, he told me, and it all worked out for him.

But the USA lost an opportunity to keep an inquisitive and courageous young academic cardiologist in the country, for the betterment of all.