The through line: a plan is only a plan once its numbers are defended
The paralysis
You have read thirty articles, notes scattered across several documents
The project feels too small (one study) and too large (impossible to finish)
The problem is not a lack of ideas. It is a lack of focus.
A prospectus is roughly one page: what you study, why it matters, how you will do it
It is a diagnostic. If you cannot state the project on one page, you cannot yet begin it.
Two questions
A: “How does livestreaming affect people?”
B: “Does the rate of chat messages per viewer differ between gaming and non-gaming streams?”
A is a career-spanning program disguised as a single question
B names what will be measured, what will be compared, and what evidence would answer it
The craft of this course is the craft of getting from A to B
What a strong question does
Specific: names the platform, the outcome, the bounded population. “How does social media influence politics?” leaves all three undefined.
Measurable: survives operationalization. “Do authentic streamers build better communities?” cannot be studied until “authentic” and “better” become observable.
Answerable within your constraints: decades of data, or hundreds of interviews, is not a semester project no matter how good the question
What a strong question does, continued
Not already definitively answered: your Chapter 4 search should show the question is open
It matters: contributes to theory, resolves a contradiction, or has applied value
Tuesday streams versus Thursday streams is answerable and still trivia
The template: does there exist + a specific variable + a relationship or difference
a bounded population + the confounds accounted for
The funnel, first passes
1: “I am interested in Twitch chat.” Too broad. What aspect? Which streams?
2: “How viewers participate in chat.” Better, still vast.
3: “Whether chat participation differs across kinds of streams.” Which kinds? Measured how?
Each pass trades breadth for depth
Scope creep is the failure mode: more questions, more theories, one more thing
The funnel, the version that ships
4: “Whether chat message volume differs between gaming and non-gaming streams”
5: “Draw a stratified sample from the 50-channel working corpus, code each message as directed at the streamer or broadcast to the room, and test whether the proportion of directed messages differs between gaming and non-gaming streams”
Version 5 fits a semester. It will not answer everything. It will answer something rigorously.
Narrowing is not weakness. It is discipline.
Three ways narrowing fails
The impossible comparison: “are Twitch viewers lonelier than non-viewers?” compares groups that differ in many ways besides the one studied
The circular question: “do popular streamers attract large audiences because people want to watch them?” defines its terms in a loop
The false binary: “is it the streamer or the game?” forces a choice the data does not require. Ask how much each contributes, and whether they interact.
Choosing an operationalization
Parasocial relationship (Horton & Wohl, 1956) is not directly observable. Three routes:
Self-report scale: captures subjective experience, invites social desirability bias
Behavioral indicators (donations, subscription length, hours watched): more reliable than self-report, but a behavior does not always reflect the bond you think it does
Language analysis of chat: naturalistic, labor-intensive, blind to viewers who feel strongly and never post
There is no right answer. You defend the proxy you chose.
Three research goals
Exploratory: “what is going on here?” Generates hypotheses. Output is patterns and themes.
Descriptive: “what does the landscape look like?” Output is frequencies and prevalence.
Explanatory: “why does this happen?” Hypothesis-driven. Output is support or disconfirmation.
Research on a topic often moves through these in order
You need not do all three. You must know which one you are attempting.
Question, hypothesis, null
A research question is interrogative: use it when exploring, or when theory makes no clear prediction, or when the literature is too contradictory
A hypothesis is declarative: use it when theory predicts, or when prior work points to a likely outcome
Behind every hypothesis sits the null: no relationship, no difference
The null is what statistical tests actually evaluate, and it is the sacred flaw again
The number that has to be defended
Statistical power is the probability of detecting an effect of a given size, if it is real
A study at 50% power has a coin-flip chance of finding a real effect
The conventional minimum is 80%, and below it a significant result is less credible: underpowered studies inflate false positives and overestimate effect sizes
In your prospectus, the number of cases is not asserted. It is defended.
Lakens on the convention
“the default recommendation to aim for 80% power lacks a solid justification.”
Lakens (2022, p. 6)
The 80% figure is a convention, not a derived optimum
Lakens treats justification by resource constraint as a last resort
The question that follows: is it defensible to run a design you know is underpowered?
Computing the required n
With an effect size from your Chapter 4 reading and α = .05:
library(pwr)# Two-group t-test, medium effect (d = 0.5), 80% powerpwr.t.test(d =0.5, sig.level =0.05, power =0.80, type ="two.sample")
Ask pwr how many cases a two-group comparison needs to have an 80% chance of finding a medium effect. The answer, n = 64 per group, is the sampling-plan justification your prospectus method section will carry.
Discussion
On Lakens (2022), “Sample size justification”:
Lakens argues the 80% convention lacks solid justification. What would a justified target look like for the study you are proposing?
He ranks resource constraints last among acceptable justifications. Is that ranking fair to the realities of a one-semester, one-coder project?
If you cannot reach the n your power analysis demands, which is the more honest move: revise the question, change the design, or report the limitation?
Five minutes in pairs, then we compare answers.
Topic, theory, and data
Three elements must align, or the study collapses
Topic is bounded by population, time, medium, or context. “Livestreaming” is not a topic. “Chat participation in the 50-channel working corpus” is.
Theory must predict patterns you can actually observe in your data
Data must be accessible, sufficient, and appropriate to the question
Misalignment example: parasocial topic, political-economy theory, chat-message data. Political economy explains industry structure; chat messages are individual expression.
Six components of a prospectus
A descriptive title hinting at the key variables
One to three focused questions or hypotheses, not five and not ten
A theoretical framework, two to three sentences, naming the theory and its relevance
A gap statement, two to three sentences, citing a few key sources
A method overview, three to four sentences: data, coding, number of cases
An expected contribution, one to two sentences
Roughly 250 words, one page, every sentence carrying weight.
The model prospectus: question and frame
Research question. Does the proportion of chat messages directed at the streamer differ between streams in gaming categories and streams in non-gaming categories, among the 50 channels in the Twitch working corpus?
Theoretical framework. Uses and gratifications theory (Katz, Blumler, & Gurevitch, 1973) holds that audiences actively select media to satisfy specific needs, including social ones.
The model prospectus: method and contribution
A stratified sample of 1,500 chat messages from the 50-channel working corpus, balanced across gaming and non-gaming streams
Each message is the unit of analysis
Three mutually exclusive codes: directed at streamer, broadcast to the room, unclassifiable
Reliability against a second coder on a 10 percent subsample, Krippendorff’s alpha
Contribution: tests whether a survey finding leaves a visible trace in coded behavior
The prospectus as pre-registration
Pre-registration commits publicly to hypotheses, methods, and analysis plan before data are collected or analyzed. The Open Science Framework timestamps the commitment.
Simmons, Nelson, and Simonsohn (2011) showed that researchers without a plan can adjust hypotheses, variables, and choices until a result reaches significance
That is the source of the term researcher degrees of freedom
Your prospectus is an informal pre-registration: the record of what you set out to do
Chapter 7: what a coder cannot see
A coder who has never watched Twitch reads “KEKW” and sees a typo, not laughter
They read thirty identical messages and do not recognize copypasta
They read “@sodapoppin same” and cannot tell agreement from a reflexive stock reply
The scheme sorts every message confidently, and the buckets do not mean what you think
Geertz (1973) called the alternative thick description: a record rich enough to capture not just what happened but what it meant in context
Three modes of attention
Casual watching: initial exposure across the range of the platform. No notes, no coding. Replace your assumptions about Twitch with exposure to it.
Focused watching: same streams, one dimension at a time. Chat behavior, viewer trajectories, host moves. Brief notes after each pass.
Analytical watching: pattern recognition across more streams. What recurs? What clusters with what? What resists description?
Familiarize, then focus, then analyze. The sequence holds for any medium.
Field notes, four kinds
Observational: what you saw. “Chat moved in bursts, near silence while the streamer talked, then a flood at every question.”
Methodological: a measurement problem. “A coder reading the log as text cannot see the question that prompted the burst.”
Theoretical: an observation connected to a framework you have read
Comparative: how cases differ. “Gaming and non-gaming chat are not the same object.”
One dated plain-text document, a note every fifteen or twenty minutes
Before Week 7
Read Chapter 7 in full and Chapter 8, “From vibes to variables”
Read the assigned article: Hayes & Krippendorff (2007), “Answering the call for a standard reliability measure for coding data”
Due next week: the Research Proposal, as a structured sentence outline
Begin your structured listening log now. The codebook is built from it.
Journal entry, 450 to 500 words, engaging both the chapter and the reading