Group assignment workshop

PSTAT197A Fall 2026

Dr Coburn

UCSB

Today

• finish discussing results from last time

• develop questions about our intake form data

• check which questions the shared data can answer

• leave with 2–3 questions and a plan for your group analysis

Assignment objective

Your task: prepare and present a descriptive analysis of the intake form responses addressing 2–3 questions or goals of your choosing.

• extend the analysis we began in class

• choose questions that are easy to state but require some work to answer

• use summaries, visualizations, and exploratory models to describe patterns among the respondents

Outcomes

The learning outcomes for this assignment are less focused on methodology:

• formulate questions and plan a simple analysis

• practice using a GitHub repo in a team project setting

• prepare a report that connects evidence to your questions

• learn one or more new-to-you data manipulation or modeling techniques

Your existing repository

Use the group repository you created on Tuesday.

• open its RStudio project and pull the latest changes

• if you have not cloned it yet, clone that same repository

• review README.md, the data and metadata in data, and the examples in scripts

• locate the report template in results/report.qmd

Keep this check brief. Students should not create another repository. Resolve individual access problems while the rest of the group begins brainstorming.

What can we describe?

• background-clean.csv: skills, comfort, coursework, and experience for 51 respondents

• interest-clean.csv: project and programming language preferences, domains, and areas of interest for 49 respondents

• join these files using response_id for analyses of the 49 respondents in both

• class-summary.csv and response-timing.csv: aggregate summaries of 55 respondents

Boundaries of the analysis

• describe the respondents represented in the relevant files

• demographic summaries cannot be linked to individual skills or preferences

• self-rated proficiency and comfort measure students’ perceptions

• an observed association does not establish a cause

Descriptive analysis can include modeling. Explore models that help answer your questions, and interpret their results in terms of these respondents.

Modeling as descriptive analysis

In class, we used k-means clustering on proficiency and comfort ratings to explore profiles of respondents.

• you are encouraged to extend that analysis or try another modeling approach

• a model can summarize several variables together or reveal patterns worth examining

• explain what the model adds to your question and how you interpret its output

Exploratory modeling is part of this assignment’s scope. Its conclusions should remain descriptive.

Starting from Tuesday’s analysis

As a group: 3 minutes

• what pattern from the class analysis caught your attention?

• what did that summary leave unanswered?

• what comparison, additional variable, or model could help you explore it?

Write down at least one possible extension.

Individual brainstorming

On your own: 4 minutes

Write down three questions you are curious about.

• consider preparation, comfort, experience, or project interests

• for each question, jot down why the answer would be interesting

• list the variables you think you would need and an analysis you might try

Keep all your ideas for now. You will check feasibility together next.

Pooling ideas

With your group: 5 minutes

1. Take turns sharing one idea at a time until everyone has contributed.
2. Record the ideas in a shared list.
3. Combine questions that address the same underlying theme.
4. Choose 3–4 candidates to develop further.

Consider both what interests you and what would add to the class analysis.

Starting with the example coursework question

Do students who have taken PSTAT 131 and PSTAT 134 rate themselves higher in programming proficiency?

Before analyzing, your group needs to decide:

• does the comparison mean both courses versus everyone else, or both, one, and neither?

• what would higher self-rated proficiency look like in the data?

• any additional question(s) that might help you understand the pattern?

Developing the coursework question

• identify completion of each course from courses in background-clean.csv

• compare the distribution of prog.prof across the coursework groups, showing group sizes

• consider whether programming comfort shows a similar pattern

• optionally explore a model relating proficiency to coursework and other preparation

Describe the association you observe. These data cannot establish that taking the courses caused higher proficiency.

Check the actual course labels before coding indicators. Keep missing course responses distinct from confirmed non-selection. Proficiency is ordered beginner/intermediate/advanced. An ordinal model is one possible extension, subject to checking the small sample and category counts.

What adds useful complexity?

• compare distributions rather than reporting just one average

• compare the same respondents across skill areas

• examine combinations of interests or preferences

• use a model to examine several variables together or summarize respondent profiles

Each additional step should help answer the question. Check group sizes before adding more subdivisions.

Possible directions

Use these as starting points if your group is stuck:

• how do programming language preferences differ across respondents’ programming proficiency levels and completed coursework?

• which application domains tend to be selected together? do those combinations differ between respondents who prefer industry versus research projects?

• how do respondents with and without a domain specialization differ in the breadth and types of data science interests they select?

• which application domains commonly appear together in respondents’ selections?

These are prompts, not assigned questions. Encourage groups to explain what they would learn and identify an extension beyond summaries already shown in class.

Extending the clustering example

What makes the respondent profiles distinct, and how much do they depend on our choices?

• compare clusters based on proficiency alone, comfort alone, or both

• try a few values of \(k\) and describe how the profiles change

• examine how scaling the variables affects the grouping

• summarize coursework or project preferences within the resulting clusters

Students should explain cluster sizes and profiles, not just show a colored scatterplot. Coding proficiency as 1/2/3 assumes equal spacing, and the comfort ratings use a different scale. Scaling and variable choices change the distances used by k-means. A difference in a variable used to form clusters is expected by construction; variables not used to form them provide additional descriptive context. Do not frame one clustering as the discovery of definitive student types.

Refining your candidates

With your group: 6 minutes

For each candidate, write one sentence specifying:

• who: which respondents will the analysis describe?
• what: which measured variables will you use?
• goal: what association, profile, or comparison will you explore?

Then explain what would make the answer useful or interesting.

Does the question fit the data?

Quick discussion

1. How does programming comfort differ by gender?
2. Does research experience cause students to prefer research-lab projects?
3. Among respondents with both sections shared, how do project preferences vary by research experience?

Which can the shared data address? What would you change about the others?

1. Not answerable from these files: gender is aggregate-only and cannot be linked to comfort.
2. Causal wording exceeds what this descriptive survey can establish. Reword as an association among respondents.
3. Feasible using the respondent-level join, subject to checking category counts. Retain or explicitly account for Unsure and No preference.

Feasibility check

Open the data and metadata: 7 minutes

• find the exact column names and files for each candidate

• inspect category labels, missing values, and group sizes

• identify needed joins, reshaping, or text processing, and any model inputs

• revise or set aside questions that the available data cannot answer

Result: select 2–3 questions to investigate. You can refine them after trying an initial analysis.

Details that affect your plan

• upper_div_courses contains ranges, not exact course counts

• project type = both means no preference

• research combines supervised data-science and other research experience

• multiple-selection variables need special care: some option labels contain commas

Use the metadata to decide what each variable can support before writing code.

Trying an idea in R

With your group: 10 minutes

Choose one candidate question and try an initial analysis.

• adapt the class script to produce a relevant summary or fit an exploratory model

• for the coursework question, start by checking course groups and proficiency distributions

• for clustering, change one choice in the class example and inspect cluster sizes and profiles

Record what you learned, what remains unclear, and what you want to try next.

What did the first attempt tell you?

Discuss your initial results:

• does the output help answer your question?

• what pattern needs a closer look?

• did sparse groups, coding choices, or model behavior create a problem?

• should you refine the question, change the analysis, or explore a follow-up?

An inconclusive result can still help you choose a better next step.

Your analysis plan

With your group: 6 minutes

For each selected question, record:

1. Exact variables, files, and respondents included
2. Data preparation needed
3. Summaries, plots, or models to try and what their output would tell you
4. Key choices to check, such as denominators, coding, scaling, or number of clusters
5. One limitation you will need to discuss

Save your notes in scripts/analysis-plan.md in the group repository.

A second pair of eyes

Exchange one question with another group: 4 minutes

Explain your question and proposed analysis in about a minute.

The other group should ask:

• is it clear what you want to learn?
• can the available variables answer the question?
• will the proposed summaries or models actually answer it?
• is the scope manageable for this assignment?

Use the feedback to revise your plan.

Dividing the work

With your group: 5 minutes

• assign a lead and a reviewer for each question

• agree on script filenames and who will edit the shared report at each stage

• set an internal deadline for initial results and a time to review them together

• choose a communication channel and arrange your next meeting

Record these decisions in your analysis plan and push it to GitHub.

Before you leave

Your group should be able to show:

• 2–3 descriptive questions grounded in the intake form data

• an analysis plan with variables, proposed methods, and limitations

• named responsibilities and internal deadlines

• the plan saved and pushed to your existing group repository

If you finish early, extend your initial analysis or try a second approach to the same question.

Deliverable

Your submission consists of work committed and pushed to the group repository, including:

• an updated results/report.qmd containing your write-up source

• a rendered results/report.html

Please push final commits by Friday, October 16, 11:59 p.m. Pacific.

There is a 24-hour grace period. Work submitted after Saturday, October 17, 11:59 p.m. Pacific will not receive review.

Resources

• lab 2 and in-class analysis: scripts/week2-surveys.R

• survey metadata and the assignment README.md

• course staff office hours and email

• MDSR ch. 4 on data wrangling

• RStudio cheatsheets and tidyverse documentation