It’s good to start a new year getting the basics right. It’s the same with methods; important not to slip into using common errors.The recent Xmas BMJ paper which showed the most common stats/methods errors is a great place to start 1/7
#MethodologyMonday

bmj.com/content/379/bm…
The BMJ stats editors highlighted the top 12 most common stats errors they come across. They are summarised in a neat infographic 2/7
All are important,but a couple particularly resonate. One is “dichotomania” (the term coined by Stephen Senn for this) where a perfectly good continuous measure eg blood pressure or weight is then arbitrarily dichotomised into two categories - good/bad; high/low etc 3/7
Often there is a clinical rationale put forward for the dichotomised groups but this needlessly throws away loads of precision & also requires higher sample sizes than if you had kept the base measurement 4/7.
See here
bmj.com/content/332/75…
Another is the issue of adjustment for clustering. It is now 25 years since @GrimshawJeremy & I published our BMJ editorial on the need for improvement in cluster trials and it is disappointing that a quarter of a century on, this remains an issue. 5/7
bmj.com/content/317/71…
The basic problem is that data which are clustered are not independent, and if you assume they are independent you can get spuriously significant results. You must adjust the standard errors to account for this. A review of recent methods is here 6/7
trialsjournal.biomedcentral.com/articles/10.11…
However, all the identified errors in the BMJ paper are easy to get right if we take the time to learn. Let’s make it our collective new year resolution not to add to these in the coming year. 7/7

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Marion Campbell

Marion Campbell Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @MarionKCampbell

Dec 12, 2022
Following the paper noted this week to have just added capital “T”s to a graph to depict standard errors 😱🤯, a short note on the importance of accurate data visualisation in any research report … 1/8
#MethodologyMonday
This was the tweet & thread which highlighted T-gate. There are lots of other issues with that paper, but data visualisation is a core element 2/8
The paper had attempted to use a #DynamitePlot (sometimes known as a Plunger Plot) to display the data. Even without adding T’s there are major issues with dynamite plots and frankly most statisticians would like them consigned to history! 3/8
Read 8 tweets
Nov 28, 2022
We are all being rightly encouraged to be #efficient in our trial design & conduct. Efficiency comes primarily through design choices … whether classic or more modern efficient designs … a few reflections below 1/7
#MethodologyMonday
A #crossover design can be highly efficient. Each person acts as their own control removing large element of variation, making the design more powerful. The outcome needs to be short term however & the intervention can’t have a long-standing effect 2/7
bmj.com/content/316/71… Image
This is particularly the case when a cluster design is also in play. A #ClusterCrossover design can majorly reduce the sample size requirements compared with a std cluster design. A good primer on this was published by @karlahemming and colleagues 3/7
bmj.com/content/371/bm… ImageImage
Read 7 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!

:(