Manifest, Latent, and the Move to Variables

Today’s Agenda

• Manifest and latent, and why the split matters
• The edge-case log, and the rules it becomes
• The gap between a field note and a variable
• Four ways operationalization fails

Tonight

Week 7 · Chapters 7 and 8

• Your notes become instructions somebody could follow
• The argument underneath the evening
  • Reliability is designed in while you are watching
  • It is not repaired afterwards
  • By the time two coders disagree, prevention has passed
• Due tonight: the Research Proposal

Where the Codebook Comes From

• You have a field-notes document
  • Observations, uncertainties, and cases that resisted
  • Those three are the raw material, in that order
• Nothing should be a desk idea
  • If you did not watch it, it does not belong
• Tonight converts the record into instructions

Manifest Content

• The surface: features explicitly present
  • Identifiable with high agreement between coders
• What that looks like in a chat log
  • An @ mention, or under ten characters
  • A known emote token, or a word count
• Countable almost mechanically, given one clear rule

The split sits at the center of codebook design (Krippendorff, 2018).

Latent Content

• The underlying meaning, which takes interpretation
• What that looks like here
  • Friendly or hostile? Aimed at the streamer or the room?
  • Is “first” a genuine claim or a running joke?
  • Is the mood celebratory or restless?
• Two careful coders can disagree in good faith

That is the definition of latent, not a sign somebody was careless.

Why Immersion Makes It Codeable

• Latent coding needs an interpretive framework
  • Built by watching rather than by defining
• Twenty hours on live Twitch teaches the pattern
  • A wall of one emote after a big play is celebration
  • Learned by watching it happen until it was obvious
• Immersion turns an impression into a defensible judgment

The Edge-Case Log

• Some observations resist every category
  • A message made entirely of emotes, no words
  • Copypasta, pasted by regulars and written by none
  • A language you do not read, or an obvious bot
  • A joke for the room wearing the streamer’s name

Log the case, what makes it ambiguous, and how a codebook could handle it.

Edge Cases Become Rules

• Not exceptions to wave past, but raw material
  • Chapter 6’s model prospectus carried “unclassifiable”
  • The log is how you learn what lands in it
• A codebook without one pays during coding
  • That is when handling them costs the most
  • A codebook with one has already decided

Knowing When to Stop

• New streams stop surprising you
• You can name three to five dimensions that matter
• The log has enough entries to write rules from
• Your notes confirm rather than discover

This is saturation again, from Chapter 4. Not a claim that you have seen everything, a claim that you have seen enough.

When Observation Changes the Question

• Observation confirms the question, or complicates it
  • Confirmed: the distinction is real and worth measuring
  • Complicated: much of chat is aimed at another viewer
• Then the two-way split has a third term
• A distinction found now is a category

Found after coding, it is a reason to start again. This is the cheapest moment.

Reliability Starts in the Notes

• You are doing two things while you watch
  • Identifying candidate categories
  • Identifying where two coders will disagree
• Notes that mark your own uncertainty
  • They are the earliest record of the reliability problem

Reliability Starts in the Notes (cont.)

• Most codebook failures are boundary failures
  • Underspecified decision rules, not bad categories
• The shape of it
  • You code a comment “hostile”
  • A second coder calls it “sarcastic”
  • The boundary between them was never drawn

The Standard This Meets

“Conclusions from such data can be trusted only after demonstrating their reliability.”

Hayes and Krippendorff (2007, p. 77)

Read that as a precondition rather than a preference. Without demonstrated reliability the findings do not count for anything at all.

One Standard Measure

• They want one measure, not a menu
• What does standardization buy a field?
• What does it cost a study with an unusual unit?

The Field Note That Is Useless

“Chat in the art stream felt calmer and more conversational than chat in the gaming stream. People talked to each other, not just to the streamer.”

A real observation, earned by watching, and useless to a coder. “Calmer” is not a category anyone else can apply, and “more conversational” is not something you can count.

The Operationalization Gap

• The variable: the target of a message, who it is aimed at
• A vague attempt: “code each message by who it is for”
• Then a coder meets a real message
  • “same.” “LULW.” “no way that just happened.”
  • Each could be the streamer, the room, or another viewer
• Named rather than operationalized

The Better Attempt

• Directed at the streamer: an at-mention of the channel
  • Or second-person address answering what they just did
• Directed at another viewer: an at-mention, or a reply
• Broadcast to the room: no specific addressee
• Unclassifiable: no criterion settles it

Observable criteria, defined categories, two coders in the same place.

Conceptual Definitions

A conceptual definition says what the variable means in the abstract. It draws on your reading and answers one question: what is this meant to capture?

The intended addressee of a chat message, reflecting whether it functions as participation with the streamer, with the chat collective, or with one other viewer.

Clear about the idea. Not yet a recipe.

Operational Definitions

• The recipe: exactly what a coder does
  • Detailed enough for a stranger to repeat you
  • The at-mention, second-person and reply rules
• Conceptual tells you what it is for
• Operational tells you what counts as evidence

Write the two side by side for every variable.

Four Failures, Conflating Concepts

• Community measured by counting messages
  • A large channel’s chat can be shallow and anonymous
  • Thousands reacting in parallel, never to each other
• Volume is activity, not community
• The fix: defend volume as a narrow proxy, or measure community

Four Failures, the Unjustified Proxy

• Stream quality measured by viewer count
  • It reflects discoverability and time of day
  • It reflects the popularity of the game, and momentum
  • None of that is quality
• A proxy is not forbidden, an undefended one is

Four Failures, Oversimplification

• Chat engagement is not only how many messages appear
  • It is who is talking, and whether to each other
  • It is also in what spirit
• A message count keeps one dimension
  • The rest is discarded silently
• The fix: measure the others, or narrow the claim

Four Failures, the Unmeasurable

• “Sincere if the viewer genuinely means it”
  • Sincerity defined by nothing a coder can observe
• The fix: use observable indicators
  • The best surface evidence for what you cannot see
• Each failure breaks the same promise, in its own way

The Promise You Are Making

• Somebody else could follow your recipe
  • And measure what you measured
• The medium changes and the act does not
  • “This article feels sympathetic” becomes a frame variable
  • “This ad feels fear-based” becomes a coder’s category

Where This Points

• A precise definition can still be bad measurement
• Reliability is consistency
  • Do two trained coders applying it independently agree?
  • If not, nothing downstream repairs that
• Kappa and alpha quantify it, correcting for chance

Next week is that machinery, and the codebook built to satisfy it.

Checking Your Proposal

• Does every variable have both definitions?
• Is each category observable, or does one read minds?
• Which variables are manifest, and which latent?
• Where will a second coder disagree with you?

Those expected disagreements are next week’s decision rules, so write them down tonight.

Before Week 8

• Read Chapter 8 in full
  • Levels of measurement, and the codebook
  • The assigned article: Lombard, Snyder-Duch, & Campanella Bracken (2002)
• Due next week: Definitions Practice
  • Conceptual and operational pairs for your variables
  • Bring three contested categories and a rule for each
• Write a journal entry, 450 to 500 words

Questions?