Manu Chevalier

Ex-academic data coach helping researchers and teams build confident R skills, using AI to assist, not replace you. Founder @DataSharpAcadem. Here to help.

[Manu Chevalier @ChevalierManu]

Hi everyone, I'm dusting off this account to embrace my new role as a data teacher and founder of @DataSharpAcadem

With DSA, I aim to help people who work with data move from messy scripts and guesswork to clear, repeatable analyses in R.

More info at https://t.co/NjJHbgsvWO

[Manu Chevalier @ChevalierManu]

Stop blaming your tools. Start by checking your data.

I’ve seen people spend hours rewriting code, changing packages, or asking AI for help…

for problems that were quietly sitting inside the dataset.

Messy data create fake complexity.

=> Fix the data, not the pipeline.

[Manu Chevalier @ChevalierManu]

The vast majority of coding problems come from:

* inconsistent formatting
* hidden missing values
* broken variable types
* messy real-world data

You probably know how to code.

But are you equipped to deal with the real complexity of data analysis?

[Manu Chevalier @ChevalierManu]

However, data curation IS integral to data analysis.

That's why it is a major focus at Datasharp Academy.

https://t.co/RRBaLT5aTV

[Manu Chevalier @ChevalierManu]

One of the biggest shocks in data analysis:

Real datasets are NOTHING like tutorial datasets.

Inconsistent categories.
Broken dates.
Missing metadata.
Mixed formats.
Unexpected NAs.

Many people think they are bad at coding.

In reality, nobody prepared them for messy data.

[Manu Chevalier @ChevalierManu]

Suddenly, everything you try seems to fail, even though it worked before.
So you look for another method, another package.

But in most cases, the issue stems from your data.
There is something somewhere that is not what you think.

Sometimes, we create the complexity ourselves.

[Manu Chevalier @ChevalierManu]

Messy data create fake complexity.

If adding 1 and 1 gives you 11, don't blame your tool.
Your data probably need some love.

A dataset with inconsistent categories, mixed formats, or duplicated labels can completely distort your understanding of what is happening.

[Manu Chevalier @ChevalierManu]

Most “coding problems” are actually data problems.

Not the algorithm.
Not the package.
Not R/Python.

Your data!

Wrong types.
Inconsistent strings.
Hidden spaces.
Duplicate categories.
Unexpected NAs.

"1" ≠ 1
"Blue" ≠ "blue"

The code is often just reacting to the chaos.

[Manu Chevalier @ChevalierManu]

That’s what we focus on at DataSharp Academy.
https://t.co/RRBaLT5aTV

[Manu Chevalier @ChevalierManu]

AI can already help you write code.

The real challenge is learning how to look at your data:
- spotting suspicious patterns
- questioning assumptions
- understanding what the dataset can actually support

That skill is built through experience, not magic.

[Past Global Changes @PAGES_IPO]

The @PAGES_ECN has launched a new mentorship program!

Applications are now open for:
🔹 Mentees seeking guidance and support
🔹 Mentors willing to share experience and advice

📅 Application deadline: 14 June 2026

Read more 👉https://t.co/fOoLC7DjSI

[Manu Chevalier @ChevalierManu]

Good data analysis is less about knowing methods and more about knowing where to direct your attention.

That skill is mostly pattern recognition built through experience.

Start looking at your data differently.

[Manu Chevalier @ChevalierManu]

Two people can look at the same dataset and draw completely different conclusions.

Not because one is smarter.
And not because one knows more advanced methods.

Often, it simply comes down to what they notice.

[Manu Chevalier @ChevalierManu]

@CMMwine thanks @CMMwine !

[Manu Chevalier @ChevalierManu]

3/
Before running models:

* compare variables
* inspect missing values
* question strange patterns
* understand what the dataset can realistically answer

Good exploratory work is not wasted time.
It is where the real analysis begins.

[Manu Chevalier @ChevalierManu]

1/
One thing that took me years to accept:

Good data analysis is often surprisingly slow at the beginning.

Not because strong analysts are inefficient.
But because they spend time understanding the dataset before deciding what to do with it.

[Manu Chevalier @ChevalierManu]

2/
When I rush early, I usually pay for it later:

* weird outputs
* meaningless models
* analyses built on poor assumptions

Slowing down early often does make everything faster later.

[Manu Chevalier @ChevalierManu]

https://t.co/RRBaLT5aTV

[Manu Chevalier @ChevalierManu]

Most people approach data analysis backwards.

They first ask:
“What method should I use?”

But strong analysis starts earlier:
- understanding the dataset,
- spotting patterns,
- questioning assumptions,

EDA is the foundation of the analysis itself.
Don't skip it.

[PAGES ECN (PAGES Early-Career Network) @PAGES_ECN]

📢 Applications are now open for the #PAGESECN Mentorship Programme!
🌍 One-on-one mentorship for paleoscience ECRs
🤝 Career guidance, networking & skill development
🧭 Connect with the global PAGES community
📅 Deadline: 14 June 2026
Apply here https://t.co/Omz5ph43LX

[Manu Chevalier @ChevalierManu]

3/ Sometimes, a linear model or a PCA already tells you enough to guide the next step.

Simpler approaches often make patterns easier to see.

Complexity has its place.
But using it too early often masks problems instead of solving them.

[Manu Chevalier @ChevalierManu]

1/ One of the most common mistakes in data analysis is jumping to complex solutions too early.

AI made this even easier. Sophisticated tools are now only one prompt away.

[Manu Chevalier @ChevalierManu]

2/ But a fancy model won’t magically compensate for poor understanding of your dataset.

People spend hours tuning models before understanding:
- the structure of the data
- the variables
- the assumptions
- or what the dataset can realistically answer