Variables and Hypotheses

Today

This is the translation step: taking an idea you believe and turning it into something a computer could check.

It sounds like the mathematical part of the course and it is not. Nothing today involves arithmetic. It involves being specific, which is harder.

You leave with a hypothesis or a research question written down, with its parts named.

From a lens to a test

• Last time: theory tells you what to look for and what would count as evidence
• A lens on its own cannot be right or wrong, because there is nothing to check it against
• Theories become checkable through variables
• A variable is a measurable concept that can take different values
• Everything today is about getting from something you believe to something you can count

The independent variable

• The independent variable is the proposed cause
• In an experiment it is what you manipulate. In a study like yours it is what you measure as the predictor.
• In livestreaming research it might be:
  • the stream’s category type, gaming or non-gaming
  • the streamer’s tenure on the platform
  • the stream’s concurrent viewer count

The dependent variable

• The dependent variable is the outcome, the thing you think gets influenced
• In livestreaming research it might be:
  • chat messages per minute
  • viewer retention across a broadcast
  • the rate of subscriptions and donations
• If you cannot say which of your two variables is doing the influencing, you do not have a hypothesis yet. You have a pair of interesting things.

A worked hypothesis

In non-gaming streams, a higher proportion of chat messages are directed at the streamer than in gaming streams.

• Independent variable: category type, gaming or non-gaming
• Dependent variable: whether a message is aimed at the streamer or at the room
• Both are things you can actually see in the dataset you already have

What testing it actually looks like

• Classify streams by category, gaming or non-gaming
• Code a sample of messages for whether each one addresses the streamer
• Run a statistical test to see whether the predicted relationship holds
• Notice how much of that is coding rather than statistics
• The codebook is where a hypothesis gets its teeth, and that is Chapters 7 and 8

The same structure, a different domain

News stories framed around individual human interest will generate more social media engagement than stories framed around systemic data.

• Independent variable: frame type
• Dependent variable: engagement
• Different topic, identical logic

Learn the structure once, carefully, and it carries to anything you study later.

Mediators: how and why

• A mediator explains how an independent variable affects a dependent one
• It is the mechanism sitting in between the two
• If category affects the rate of directed messages, the mediator might be the perceived pace of the stream
• Non-gaming streams may feel more conversational, and that conversational feel may be what invites viewers to speak to the streamer directly

Moderators: under what conditions

• A moderator changes the strength or direction of a relationship
• It answers “under what conditions?”
• The category-to-directed-message relationship might be moderated by audience size: holding on small streams and weakening on large ones
• A moderator is not a nuisance to be controlled away. Often it is the finding.

The ladder

• “X relates to Y” is where a first project starts, and that is a real result
• “X relates to Y, through this mechanism” adds a mediator
• “X relates to Y, through this mechanism, under these conditions” adds a moderator
• You do not need all three rungs for a good White Paper
• Knowing the ladder exists is what stops you claiming rung three from rung one

Choosing a theory: the first three steps

1. Get specific about the question. “Livestreaming and community” is too broad. “Does the proportion of chat messages directed at the streamer differ between gaming and non-gaming streams?” is something you can act on.
2. Map the question to a theoretical domain. Group processes point toward social identity or uses and gratifications. Perception effects point toward cultivation, agenda setting, or framing.
3. See how others studied similar questions. Repeated use of one theory in a domain is a signal it has been productive there.

Choosing a theory: the last two

4. Let the theory produce the prediction. Derive the hypothesis from the theory, rather than picking a theory afterwards to decorate a hunch you already had.
5. Design measures for the concepts. Turn abstractions like identity or salience into variables somebody could actually record.

Step 5 has a name, operationalization, and it is most of the next two chapters.

Hypothesis or research question?

• State a hypothesis when your theory is precise enough to predict a direction: more, less, higher, lower
• Ask a research question when the theory supports expecting a difference but not which way it goes
• A research question is not a weaker hypothesis. It is an honest one.
• What you must not do is ask a question, look at the data, and then write it up as though you had predicted the answer

Why the hypothesis comes first

• HARKing is Hypothesizing After the Results are Known
• You look at data, spot a pattern, then write as though you called it in advance
• The p-value still prints. It just no longer means what it says, because the hypothesis came out of the data instead of preceding it.
• Writing your prediction down before you look takes five minutes and protects every inference that comes after it

Before next time

• The Annotated Manuscript is due this week, 25 points
• CITI Ethics Certification is due this week, 25 points. If you have not started it, do that tonight.
• Write your working hypothesis or research question into your project repository, with the independent and dependent variables named
• Read Chapter 6 before Thursday
• Thursday is research questions: when a question is the honest choice, and how to write one that can actually be answered