Down and Out, revisited

· ·

An honest re-examination of my 2023 MA thesis, built in the open, and the accessible story that will come out of it. This page changes as the work does. Every change is dated at the bottom.

Why I'm looking again

In 2023 I completed my Masters degree in sociology. What I didn't know then and what I know now is that you can get things wrong and still get a degree. Statistics were never my strong suit, and while rereading my thesis reaffirms that for me, it also makes me wonder if I'd not muddled through the numbers in grad school, would I be where I am today? Who knows. However, I do know that questions about structural inequality, the role of institutions, the people who move through them, and the ways those things interact are still ones that I think about regularly (and in my opinion are still very worth asking).

As such, with the help of Claude, I hope to review my thesis, tell the story of the process, make the information more accessible, and figure out where I went wrong (and how I might change things from here).

What I asked in 2023

My thesis, Down and Out: Support Services, Satisfaction, and the Impacts of Marginality Among Canadian Youth Experiencing Homelessness, asked a simple question: are young people experiencing homelessness all served equally well by the services meant to help them? And if not, do marginalized identities, such as being Indigenous or LGBTQ2S+, help explain the difference?

To answer it, I used Canada's 2015 national youth homelessness survey. It reached 1,103 young people through 57 agencies across the country. Youth who had used services such as shelters, food banks, health and mental health care, and employment programs were asked how helpful those services had been. I used regression models to look for patterns in their answers.

The study at a glance

  • 1,103young people surveyed
  • 57agencies that ran the survey
  • 5service areas youth rated
  • 2015when the survey ran
The service areas were employment and drop-in programs, food banks, health care, mental health and addictions services, and shelter and housing. Sources: the thesis and the survey's national report.

The original thesis is on Atrium, the University of Guelph's repository. It stays as it was. This page doesn't replace it.

What I can and can't claim for now

I'm starting from what was published: the thesis, its tables and appendices, and the survey's national report. I don't currently have the survey data itself. Working from the published material, I can:

  • restate the published numbers accurately, recomputed from the thesis's own tables
  • read the published results correctly, in plain language
  • sort every claim the thesis makes: supported, unsupported, contradicted, or unknowable from what was published
  • describe the study I'd design now
  • explain the statistical ideas with invented, clearly labelled data

What the published material can't do is support corrected findings, or confirm anything that would need the raw data. Where something is my best explanation rather than something I can check, I'll say so.

What about the data?

The survey data were made available to me for the thesis, and I don't currently have access to them. That could change. I may be able to get access again, or find other data that let me ask similar questions. Either would open up new work, and this page will say so if it happens. This first phase doesn't wait on it.

How I'm working

Each question goes through the same five steps. The mark on each step says whose work it is.

How each question gets worked

  • dashed: open, unresolved
  • pen hatch: my judgement
  • pixel: Claude's review
  • solid: recorded verdict
  • ruled: documented here
  1. 1 · openOpenThe claim, quoted from the thesis, with the evidence to check.21
  2. 2 · first passMy first passI work through it from the sources myself.0starting here
  3. 3 · reviewClaude's reviewClaude, an AI model made by Anthropic, checks my pass, raises explanations I missed, and compares it with its own earlier read.0
  4. 4 · verdictVerdictA status and a confidence level, and what it rests on: arithmetic I can re-run, a cross-check against another source, or inference.0
  5. 5 · publicPublicA plain-language version comes here.0
The number on each step is how many of the 21 questions are there now.

Every number is recomputed from the thesis's published tables by small scripts, so it can be checked.

One honest caveat: before setting this up, I read Claude's full preliminary review of the thesis, so my first passes aren't blind. I'm noting that on every item. Claude got some things wrong in that review too, and those corrections will show up here alongside mine.

The questions I'm working through

All 21 are open. I've worded them as questions on purpose; the answers come once I've done the work. They're grouped by where each one enters the work, from what was measured to how it was framed, because problems early on carry through to everything after.

Where the questions sit

What was measured
  • E01
  • E02
  • E03
  • E04
  • E05
  • E06
  • E07
  • E17
8
How the study was designed
  • E08
  • E09
2
How the results were tested
  • E10
  • E11
  • E16
3
How the results were reported
  • E12
  • E13
  • E18
3
How the results were read
  • E14
  • E15
2
How it was framed
  • E19
  • E20
  • E21
3
Every question is dashed for now. Each one takes on a new mark as it moves through the five steps, so this figure doubles as a record of progress. The IDs match the project's register. The questions themselves are listed below.
What was measured (8)
  • E01What does the response scale actually measure?
  • E02How was past suicidality coded?
  • E03How was gender coded, and who was included?
  • E04How were race and Indigeneity coded?
  • E05What does the "LGBTQ2S+" variable capture?
  • E06How were the other predictors built?
  • E07How were the combined service ratings built?
  • E17Does where the survey was given matter?
How the study was designed (2)
  • E08Could the models test compounding marginality?
  • E09Which way do the effects run?
How the results were tested (3)
  • E10How were the significance tests set up?
  • E11How well do the models fit?
  • E16Are the models comparing the same people?
How the results were reported (3)
  • E12Do the results tables agree with themselves?
  • E13Do the before-controls tests match their appendix?
  • E18What does the ranking of services show?
How the results were read (2)
  • E14What does the age result mean?
  • E15Do the data support the shelter explanation?
How it was framed (3)
  • E19How old were the data when the thesis was written?
  • E20What age range does the study cover?
  • E21Are the framing and citations accurate?

One of these questions concerns how suicide attempts were recorded in the survey. If that topic touches you personally, support is available any time: in Canada, call or text 9-8-8, the Suicide Crisis Helpline.

What's next

First up is the response scale: what youth were actually asked, and what their answers show. That's where the first interactive piece will go.

Revisions

  • 28 Oct 2026. First version: the question, the method, and the open questions. No findings yet.

Sources

  • Lathangue, S. R. (2023). Down and Out: Support Services, Satisfaction, and the Impacts of Marginality Among Canadian Youth Experiencing Homelessness (MA thesis, University of Guelph). Atrium
  • Gaetz, S., O'Grady, B., Kidd, S., & Schwan, K. (2016). Without a Home: The National Youth Homelessness Survey. Canadian Observatory on Homelessness Press. PDF
  • This project is done with Claude, an AI model made by Anthropic. Its part is labelled as its own.