EDF9850_Activity1_Part2
Watch on YouTube →
Overview
Lesa Hoffman reviews course deadlines and project planning, then recaps how multilevel models handle level-two predictors and clustered data. She emphasizes that level-two main effects explain cluster-mean variance, pseudo R-squared uses a narrower reference point than total R-squared, and ordinary least squares can produce misleading standard errors and significance tests when it ignores clustering.
Key takeaways
- Level-two predictors used as main effects explain differences between cluster means, represented by level-two random-intercept variance.
- Pseudo R-squared is calculated against a narrower level-specific variance component, so it can make an effect appear larger than total R-squared; report clearly what variance the percentage refers to.
- Ignoring clusters in ordinary least squares can produce underestimated standard errors and inflated Type I error because the analysis treats people, rather than clusters, as the relevant independent units.
- Unequal cluster sizes can make level-two slope estimates from a full individual-level analysis differ from estimates based on level-two data.
- Cluster-corrected standard errors adjust inferential uncertainty, but they do not fix a slope estimate that is itself distorted by ignoring the multilevel structure.
- Submitting a mostly complete homework before the deadline is generally preferable to losing three points for lateness over one or two unanswered questions.
Chapters
- Homework 1 is available in the homework system and due September 16; Hoffman expects students to have the material needed after this class.
- Homework 0 may show as complete without being submitted, so students should re-enter their ID number; incomplete work can still be submitted to avoid a three-point late penalty.
- The five-point project plan is due three weeks later and asks about sampling design, sample sizes, variables, and research questions.
- Students can work independently or with a partner, submitting one shared document; existing research projects are suitable, and presentation dates will be corrected and reposted.
- The class scales the activity workload back to one application with six questions after the previous set proved more demanding than expected.
- Level-two predictors as main effects, and interactions among level-two predictors, explain level-two random-intercept variance, or cluster-mean differences.
- Pseudo R-squared can look larger than total R-squared because it measures explained variance relative to the cluster-level variance pile rather than all outcome variance.
- Hoffman recommends stating the reference explicitly, such as the percentage of total variance or of cluster-mean differences explained.
- Single-level ordinary least squares ignores clustering and can inflate Type I error by assessing uncertainty as if the sample size were people rather than clusters.
- Level-two predictor slope estimates may match between single- and multilevel analyses only in perfectly balanced data; unequal cluster sizes can also change estimates.
- A standard error that is too small produces an overly optimistic p-value, making effects appear significant more often than warranted.
- Cluster-corrected standard errors address clustering-related uncertainty but do not repair incorrect slope estimates; students can use the Shiny app through CU Apps if their R setup cannot run it.
Transcript excerpt
0:00 Enough fun for now. Let the real fun begin, right? It's 5:03. Welcome back, everybody. I'm happy to see you that the the stool that I stole so I could sit comfortably during class is still here. I feel nice and tall and nice and elevated up here. It's great. Uh I've been busy putting some new stuff up for you to do. So, I wanted to make sure I pointed all of that out before we go through our tasks for today. Um homework one is now available inside the homework system. Okay, you don't have to care yet. I saw the look. Homework one is not due for 2 weeks. Let me Let me get the exact date on here. Yeah, 9 16. So, I put it up in case you want to get started. My intent is after we finish today, you should be able to do all of
0:46 it. Or if you'd like to ignore it for another week and a half, that's fine, too. It's up to you, right? I will say that in the past students have used my homeworks as a way to be productive and yet procrastinate on other things because they're very prescribed in what you have to do, and sometimes that feels better psychologically than write this paper or figure out this research plan. Like amorphous tasks are uncomfortable, but this is very very prescribed. So, if you'd like to procrastinate using my materials, knock yourselves out. Um everybody got through homework zero to get some extra credit points. Thank you for testing that. Um did anyone have any questions or concerns about what that looks like? Okay, the one thing, don't forget to re-enter your ID number to submit for real. This happens usually several times
1:33 a semester that folks forget to do that.
This is the start of the transcript. Save this video free for the whole transcript, searchable and time-coded.
Related notes
Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Lesa Hoffman.