iiinova — where your digital self keeps evolving

Christopher Jarvis @christopherjarvis.bsky.social · 17/05/2025

When I was an MSc student I made nearly all the figures for a paper just to get an acknowledgment. I’d love to meet that person now to discuss/berate them as to why they felt this was okay.

040

Christopher Jarvis @christopherjarvis.bsky.social · 17/05/2025

A challenge I faced was push back from senior authors for adding junior authors who contributed more than most of the seniors. In extreme, seniors sent emails saying well done after submission and congrats after acceptance. Juniors at the least read the paper and provided useful comments.

130

Reposted by Christopher Jarvis

Em Prestige @emprestige.bsky.social · 28/04/2025

Happy to share the publication of our work "Estimating social contact rates for the COVID-19 pandemic using Google mobility and pre-pandemic contact surveys", huge thanks to my co-authors @christopherjarvis.bsky.social @ngdavies.bsky.social @pietro-coletti.bsky.social Jantien Backer and John Edmunds

doi.org

Redirecting

1146

Christopher Jarvis @christopherjarvis.bsky.social · 28/03/2025

You could argue USA is big and diverse but then Canada looks like it varies so they thought about using a quiz to help and they start off with support and kind words www.canada.ca/en/services/...

canada.ca

What to do when someone dies - Canada.ca

Dealing with a loss can be overwhelming. Our guide provides comprehensive information, resources and guidance related to survivor benefits and the steps to take in the case of a death.

020

Christopher Jarvis @christopherjarvis.bsky.social · 25/02/2025

Variables must be named after the author in numerical order. Chris chris1 chris2 chris3 if you need to make a new one earlier on then add a letter chris1a

160

Christopher Jarvis @christopherjarvis.bsky.social · 25/02/2025

No worries. If you get really stuck I’d be happy to have a call.

010

Christopher Jarvis @christopherjarvis.bsky.social · 25/02/2025

If the values at the locations are of interest, rather than the locations then you want to adjust for spatial correlation. This is where you used random effects models that account for spatial structure. Such as SAR, CAR, ICAR, BYM models. These fall in areal or geostatistics.

110

Christopher Jarvis @christopherjarvis.bsky.social · 25/02/2025

There are a couple of things: If the randomness you’re interested is the location, which as you’re calculating distances and analysing them that’s seems to be the case then you can look at point pattern analyses. en.m.wikipedia.org/wiki/Point_p...

en.m.wikipedia.org

Point pattern analysis - Wikipedia

110

Christopher Jarvis @christopherjarvis.bsky.social · 24/02/2025

Hi @caprosser.com, the problem isn't quite clear to me. I'm not sure it's spatial. Isn't this the same as simulating height and then measuring how far all of them are from 10cm intervals then showing that some of them are nearer to a bunch of the intervals? and then correlations due to boundaries.

100

Reposted by Christopher Jarvis

Tim Morris @timpmorris.bsky.social · 17/01/2025

Stata users: if you want to make ‘kmunicate’-style Kaplan–Meier graphs, I’ve just released a Stata package on GitHub. To get it: . net from raw.githubusercontent.com/tpmorris/kmu... User feedback welcome! 1/

A vanilla KMunicate-style Kaplan–Meier graph from {kmunicate}. Key is the extended risk table showing the status of participants over time.

33415

Christopher Jarvis @christopherjarvis.bsky.social · 22/12/2024

This is a nice option that moves beyond lots of copy paste in your script and gets you closer to thinking about repeated patterns that you might eventually put in a package. #rstats

061

Christopher Jarvis @christopherjarvis.bsky.social · 18/12/2024

And congrats! Last PhD paper back with revisions that took you a couple of hours is a big achievement!

010

Christopher Jarvis @christopherjarvis.bsky.social · 18/12/2024

Definitely, once you’re used to submitting papers you realise a lot of these deadlines are arbitrary and if they’ve already sent out to reviewers and you’re asked to respond then they’re pretty invested. I found asking for a month or more is accepted with no fuss.

110

Christopher Jarvis @christopherjarvis.bsky.social · 18/12/2024

After I’ve said this, I realise it’s a the same system we use for driving tests in the UK. My motivation is that I want to cause the authors the least amount of work required for the paper to be publishable. Plus what I might think takes hours takes them weeks or vice versa.

010

Christopher Jarvis @christopherjarvis.bsky.social · 18/12/2024

I think of major as if you don’t address this then I’ll probably reject the paper. Then minor is suggestions that I’ll let pass unless you ignore lots of them. The amount of time they take is secondary to me, I think of them as importance.

150

Christopher Jarvis @christopherjarvis.bsky.social · 17/12/2024

My fav shortcut is alt+shift+k which shows you all the shortcuts.

010

Christopher Jarvis @christopherjarvis.bsky.social · 17/12/2024

Then sometimes you have loads of computers (docker images) that you want to switch on many times (containers) so you use something like Kubernetes to manage all of that for you.

010

Christopher Jarvis @christopherjarvis.bsky.social · 17/12/2024

I like this intro colinfay.me/docker-r-rep... And tugboat is here www.dmolitor.com/tugboat/ The mains bits are the dockerfile which records how you want to set your computer up (the image) and the docker-compose file which allows you to build the computer, and also switch it on (a container).

colinfay.me

An Introduction to Docker for R Users

A quick introduction on using Docker for reproducibility in R.

101

Christopher Jarvis @christopherjarvis.bsky.social · 17/12/2024

I found #docker mysterious until I thought of it like buying a laptop setting it up exactly how I need and then handing it over to another person. Then dockerhub is like the post office that takes my laptop and delivers it to others. Tugboat makes it pretty easy to do in #rstats

140

Christopher Jarvis @christopherjarvis.bsky.social · 17/12/2024

I feel like I started my #stats journey wanting to do lots of fancy multi-level, spatial, and Bayesian models. Then I'm now happy if I can produce a useful count, that I understand and can clearly explain. In many ways knowing the complex stuff just gave me permission to do simple things. #rstats

Left hand side has a complex formulate for multi-level models, the right hand side has code to count up the number of rows in a group in dplyr.

080

Christopher Jarvis @christopherjarvis.bsky.social · 12/12/2024

Here’s some proof bsky.app/profile/chri...

000

Christopher Jarvis @christopherjarvis.bsky.social · 12/12/2024

I like to leave them all blank and leave interpretation to the reader.

170

Christopher Jarvis @christopherjarvis.bsky.social · 10/12/2024

Cool how often are you using it? I find dput useful but I’m not using it a lot, so a bit unsure about installing another package. Maybe there are some use cases I haven’t thought about?

100

Christopher Jarvis @christopherjarvis.bsky.social · 10/12/2024

I was chatting with a colleague about their #PhD viva on #spatialstats and I showed them this diagram that I made as as part of my thesis to help me understand the building blocks of #spatial models, especially for #inla. #statssky. pg 254 tinyurl.com/3ymp26p3

041

Christopher Jarvis @christopherjarvis.bsky.social · 09/12/2024

Today I learned that the R in #RShiny is doing a lot of the heavy lifting in that hashtag. #shinyisnotjustR #Rstats

030

Reposted by Christopher Jarvis

Norm Matloff (你有冇諗清楚呀?) @matloff.bsky.social · 09/12/2024

Knuth, one of the greats in CS history, famously said, "Premature optimization is the root of all evil...in programming." Only a small % of R apps need to be efficient. But they all need to be correct, and often loops etc. are clearer and thus more likely to be correct. Anti-loop mania is bad for R.

192

Christopher Jarvis @christopherjarvis.bsky.social · 09/12/2024

Do you have any publicly available that you use on github?

100

Christopher Jarvis @christopherjarvis.bsky.social · 09/12/2024

Hey James, yep I do! Really helpful. Before that I was using shell scripts to do the same.

210

Christopher Jarvis @christopherjarvis.bsky.social · 09/12/2024

Welcome to Bluesky Nick! Sounds similar to anxiety which would be energy draining. I wonder if adults that aren’t constantly doing social evaluation have more energy.

010

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

I'm a novice at package dev and this was an idea I had this morning. Certain that others could improve it loads but this already does something that I've missed since switching from Stata 12 years ago. bluesky seems full of those knowledgeable people. Help welcome. github.com/jarvisc1/qex...

github.com

GitHub - jarvisc1/qexplore: R package for quickly exploring data.

R package for quickly exploring data. . Contribute to jarvisc1/qexplore development by creating an account on GitHub.

010

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

This isn't trying to replace anything, I'd see this as sitting in that sweetspot where you need to really quickly find the rows that meet a certain condition and only want to see certain columns. Then you need to switch quickly to some other condition and columns. Throw away console code.

110

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

qtab: Similar but you can do tables with one, two, or even three variables. if_ and in_ work the same but we also have by_ to make 2x2 tables for each category of a third variable. The input is a data.frame and the output is a tabyl. This means it can be combined with #gt and #flex.table

100

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

qlist: Tell the function what variables you want and then filter using the if_ arg and pick specific rows using the in_ arg. Input and output are data.frames so it's tidyverse compatible. The if_ command should be compatible with anything you could send to filter in dplyr.

100

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

The idea is to speed up having to type multiple commands with #dplyr when trying to explore data and to give something with more input than skimr. I like being able to tabulate and print based on subsets of my data. There are two commands qlist and qtab for listing and tabulating your data.

100

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

bluesky is inspiring! I think I've made my first #R package #rstats #eda github.com/jarvisc1/qex.... Inspired by tidyplots@jbengler.de and a chat with @pete-mck.bsky.social bsky.app/profile/did:.... I thought why not try to bring what I like about #Stata and #data.table to the #tidyverse.

github.com

GitHub - jarvisc1/qexplore: R package for quickly exploring data.

R package for quickly exploring data. . Contribute to jarvisc1/qexplore development by creating an account on GitHub.

292

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

Absolutely same as relative versus absolute effects. Double your risk of death sounds really bad! But from 1% to 2% maybe not as bad as from 40% to 80%. Though at a population level 1% to 2% would still be devastating. Fun chatting with you.

010

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

Then to your point, the other big thing to consider is the magnitude of scale on which your uncertainty is. So uncertain R number has huge implications because it’s on an exponential scale but uncertain fatality maybe not as bad.

100

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

It is yeah! I’ve also noticed that causal inference and DAGs are pretty similar to to transmission models expect their causal mechanism is germ theory from biology so it’s a really strong mechanism and less hand wavy.

110

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

A friend called all their objects after themself! bob1 bob2 bob3. I loved and hated their code. I use a style of x_df because I’ll want to group the dfs by name then if I’m making a table or plot it will be tab_x and plot_x so I can find all tables and plots and still link to relevant dfs.

000

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

Make sense because they often come from a physics background, so they’re not shy about the underlying maths at all. Whereas a subset of Stats has become hugely accessible so you get a lot of non-maths statisticians. Which is fine I think. Glms can get you very far.

110

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

They’d still use glm but only if they exactly needed glm. I realise they were not restricted by learning a small subset of models. They mostly thought in terms of germ theory, inputs then fitting and outputs.

110

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

When I moved into infectious disease modelling it was fascinating, for them the answer would be yes it’s trivial. I realised they were more like traditional statisticians than most current statistics PhDs. They built models from scratch and had a range of hand-written algorithms for fitting.

121

Christopher Jarvis @christopherjarvis.bsky.social · 08/12/2024

I’ve always thought of a confidence interval as the time period between me saying what a confidence interval is and anyone else then telling me it’s something else. I’m usually pretty confident during that brief interval. #statssky #episky

070

Christopher Jarvis @christopherjarvis.bsky.social · 07/12/2024

Ahh I wasn’t clear. Don’t worry, I liked your presentation.

100