Manifest and Latent Content

Today’s Agenda

• The two layers of meaning in a message
• Sorting your own notes into easy and hard
• Sampling what you observe
• Edge cases, and the rules they become

The Argument You Will Have

• You and a second coder will disagree
  • Same chat message, two readings
  • Neither of you is wrong, exactly
• Today is about seeing it coming
  • Build the codebook that settles it in advance
  • Not at eleven at night in November

You leave having sorted your field notes into the easy half and the hard half.

Two Layers in Every Message

Krippendorff, 2018

• The split sits at the center of codebook design
• Manifest content: what is explicitly on the surface
• Latent content: the underlying meaning
  • It takes interpretation to reach
• Most first codebooks treat a latent variable as manifest

Manifest Content

Features any trained coder can identify with high agreement:

• Does the message contain an @ mention?
• Is it shorter than ten characters?
• Does it contain a known emote token?
• How many words is it?

These are countable almost mechanically, and the work is writing one clear rule.

Latent Content

Features you cannot read off the surface:

• Is the message friendly or hostile?
• Is it directed at the streamer, or performing for the room?
• Is “first” a genuine claim or a running joke?
• Is the mood of chat celebratory or restless?

Two careful coders can disagree about every one of these in good faith.

The Argument for Watching

• Latent coding needs an interpretive framework
  • It turns an impression into a defensible judgment
  • That framework gets built by watching
• Twenty hours of live Twitch teaches the pattern
  • A wall of one emote after a big play is celebration
  • Learned by watching it happen, repeatedly
• Immersion is how latent content becomes codeable

Sorting Your Notes

• Mark every observation M or L
  • Manifest items are variables you could write today
  • Latent items will need decision rules
• A workable codebook mixes both
  • Two or three manifest measures, one that takes judgment
• All latent, and your reliability check is going to hurt

Sampling Your Observation

• You cannot watch all of it, and need not
• What you need is a structured sample
  • It should reflect the range of the thing you study
• Ten large competitive-gaming channels teach you one thing
  • About everything else they will actively mislead you

Match Your Source’s Range

• The class dataset spans a deliberate range
  • 50 channels across the volume distribution
  • Gaming and non-gaming, crowded and nearly empty
• That range is the principle, not the assignment
  • Span whatever source you settled on with the librarian

Match Your Source’s Range (cont.)

• On Twitch: big channels and small, gaming and art
• On a subreddit: busy threads and dead ones
• On a news site: the front page and the unlinked pages
• Sample across time as well
  • A feed at peak and in a quiet hour are different rooms

Coverage Beats Volume

• Twenty streams spanning the variety beat fifty alike
• The same principle keeps returning
  • The wide literature search sees the whole conversation
  • Statistical sampling later in the term does the same
• Representativeness is a discipline at every stage

When Channels Are Live

Figure 7.1, the collection week

• Eight channels across the distribution, November 18 to 24
• Large channels are live daily, for long stretches
• Small channels appear once or twice all week
  • A random two-hour window catches the big ones easily
  • It catches the small ones only if you place it
• Scheduling your observation is a sampling decision

The Edge-Case Log

Some of what you observe resists every category you are tempted to draw. Those moments are the single most useful thing observation produces.

Keep an edge-case log beside your field notes. For each case: log it, say what makes it ambiguous, and list the ways a codebook could handle it.

Four Edge Cases

• A message of pure emotes, no words
  • Directed, broadcast, or a different object entirely?
• Copypasta: a block every regular pastes unread
  • A message, and not written by that viewer in any normal sense
• A language you do not read
• An obvious bot, or a joke wearing the streamer’s name

Edge Cases Become Rules

• These are the raw material of decision rules
  • The model prospectus carried “unclassifiable”
  • The log is how you learn what lands in it
• A codebook without one pays later
  • It meets these messages first during coding
  • A codebook with one has already decided

Knowing When to Stop

Four signs immersion has done its job:

• New streams stop surprising you
• You can name three to five dimensions that matter
• The edge-case log has enough entries to write rules
• Your notes repeat themselves rather than adding

This is saturation again, the same idea the literature search used.

What You Walk Away With

• Conceptual clarity about your variables in this medium
• Candidate categories for each one
• The beginnings of the decision rules for hard cases
• Enough context to spot a surface feature about to mislead

Those four are the raw material of a codebook, which is what you build next.

Before Next Time

• Finish your notes and edge-case log
  • Three or more sessions, big and small, gaming and not
  • Mark each entry manifest or latent
  • Star the three a second coder will fight you on
• Write two or three candidate questions
  • Grounded in what you saw, not what you assumed
• Read Chapter 8, through how operationalization fails

Questions?