Stephanie Eckman

NLPOR workshop at COLM

ChatGPT generated image for the happy hour event after the workshop

are you attending @colmweb.org 🦙 psst, don't miss our First Workshop on Bridging NLP and Public Opinion Research tomorrow #COLM2025

🛎️ we have a great line up of keynotes, presentation, and panel discussion, followed by a happy hour

📍 room 518B

schedule and info: sites.google.com/view/nlpor20...

I know the pro version of Claude will let me load up papers, that’s the next approach I’ll take

I tried similar prompts with the Assistant in SCite, but it gave errors and didn’t produce anything. I think it does better with shorter prompts (“find references to support / disagree with this statement”)

(@scite.bsky.social exists, but I can't tag it)

When I finally got it to make citations in the style I wanted, many of them were incorrect (which is wild, when there are only 40 or so papers loaded).

In the end, it didn’t save me any time at all

From the notebooklm community on Reddit Explore this post and more from the notebooklm community

It writes OK lit review text, but I can’t get it to make usable citations. It makes those little numbers I can hover over, but it’s a ton of manual work to turn those into \cite{xyz} commands.

A lot of complaints about similar issues online: www.reddit.com/r/notebooklm...

Looking for tips on how to use the new AI tools to help with a literature review.

I was so excited about NotebookLM, but my frustration is growing. I uploaded the papers (~40) I want it to draw from, and then spent several hours trying to get it to draft a reasonable lit review *with citations*

@statstas.bsky.social — your point about totals rather than percents

It’s really funny, though not a comedy

Here’s a paper doing imputation. I read and earlier version, not the latest, and was frustrated that they didn’t compare their ML imputation method to other more traditional imputation approaches.

arxiv.org/pdf/2305.096...

This is the closest I’ve found:

arxiv.org/abs/2401.08572

How do you make a scientific case about a method’s usefulness without a proper lit review?

by synthetic samples, I mean using LLMs to generate survey responses

"imagine you are a 54 year old African American man. Please answer the following survey question..."

Can anyone point me to a paper, WP, blog post that reviews the literature on synthetic samples? Thank you

Has anyone used DC’s school lottery as an instrument to evaluate effect of pre-k on subsequent educational achievement ?

IMO, the worst problem with data analysis a in Excel is how hard it is to trace back where a result came from. Yes, there are now trace dependence and trace precedence buttons, but it still is not nearly as clear as it would be in a script file

Took us a while, but we finally put another chapter of our book "Data Collection with #Wearables, #Apps, and #Sensors" online. Chapter 4: Errors of Nonobservation: #Sampling and #Coverage. bookdown.org/wasbook_feed... Let us know what you think! @stephnie.bsky.social

They really are : “the importance…in some vague sense” IIRC

I show students the definitions of the 4 types of weights, to try to prevent them from using aw= (because that’s what the command accepts) when they have pweights. I doubt I’m successful

In case you missed it! "Interviewer Involvement in Respondent Selection Moderates the Relationship between Response Rates and Sample Bias in Cross-National Survey Projects in Europe" by Marta Kołczyńska, Piotr Jabkowski & Stephanie Eckman @stephnie.bsky.social doi.org/10.1093/jssa...

@aapor.bsky.social @floriankeusch.bsky.social @daob.nl

Fantastic thread on data quality issues with inattentive respondents.We can’t weight our way out of these problems if the demographics are provide are wrong

Debatable — what do you mean by “very well”?

Sort of — in R, PLmixed will give you similar results (discrimination, difficulty) to mirt. PLmixed fits “mixed models with factor structures”

because they're economists 😉?

search results for "aapor.org conference"

Searching for the dates of the next AAPOR conference, I typed in "aapor.org conference." None of the top hits were helpful --previous years, archived site

I think we could use a little Search Engine Optimization, @aapor.bsky.social

In the meantime -- include 2024 in your search (or "79th" 🙄🤪)

New Article Alert! "Interviewer Involvement in Respondent Selection Moderates the Relationship between Response Rates and Sample Bias in Cross-National Survey Projects in Europe" by Marta Kołczyńska, Piotr Jabkowski & Stephanie Eckman @stephnie.bsky.social doi.org/10.1093/jssa...

Yes — some colleagues want to see an intercept slope version before they’ll trust the results

well, right this minute I'm trying to show that IRT models are just multilevel logit models, which I'm told is true, but I cannot get the code to work

any help appreciated!

Bluesky, LET'S DO THIS!

#stats #statistics

Ooo, that would be a good thread — statistical programming issues that have stolen hours of our lives