Please enable JavaScript.

Coggle requires JavaScript to display documents.

PSYC232 Test 3, Regression: predicting a score using values from one (or…

- - - - predicted romantic interest = intercept + attractiveness of date + error
        
        Additional information
  - - - e.g.
        
        "A one-unit increase in perceptions of a person being attractive/sexy was associated with a .809 increase in romantic interest, controlling for gender."
    - - e.g.
- - - - R is like a correlation but for everything in the regression model. How much is the line that is made up of all of your predictors associated with the outcome?)
        
        If you have only one predictor, then the model R is the same as the correlation r for those two variables
      - R2 is the amount of variance in the outcome that it explained by the regression model.
        
        Model error: after modelling the covariance of the predictor/s and the outcome, how much variance in the outcome is left over
        
        "Error" is the amount of variance not explained by our model. (e.g. what else predicts romantic interest?)
        
        Calculating error: Calculated by "1 - R2"
        
        In the image, 49.5% of the variance is error (1-0.505 - 0.495)
        
        variance explained: after modelling the covariance of the predictor/s and the outcome, how much variance in the outcome is known
      - Adjusted R2: is a correction of the R2 for a model with more than one predictior
        
        We usually report both R2 and adjusted R2. This is because adding predictors can increase prediction due to random chance (adding predictors can never make the predicted estimate worse).
      - Model test: does your model explain significant variance in your outcome compared to chance?
        
        is the amount of variance explained (R2) statistically different from explaining zero variance?
        
        e.g.
        
        if this is significant: the predictors in this model can give you a number of someone's romantic interest in their speed-date that, on average is closer than guessing the mean
    - - Regression model labels
        
        Outcome = romantic interest (e.g.)
        
        Dependent variable: the outcome in experimental models
        
        Independent variables: measured predictors in experimental models
        
        Covariates: predictors that are controls in experimental models
      - "Intercept" / "Constant": Where the regression slope crosses the y-axis in the equation. The value of the outcome when all predictors have a value of zero ("held constant")
      - [unstandardized] Estimate = "B": The amount of change in the outcome associated with a one-unit change in the predictor variable.
        
        E.g. For every one-unit increase in a person rating another person as being attractive/sexy (on the 1-10 scale), our model predicts that their romantic interest should increase by .84 units.
        
        The "t" and "p" test whether the estimate for this predictor is significantly different from zero.
        
        Standardised estimate: The amount of standard deviation unit change in the outcomes associated with one standard deviation unit change in the predictor variable.
        
        This is an effect size for the predictor because it is on a standardised scale (just like a correlation).
        
        (written up, apa style) check notes later for how to do this
        
        still better than guessing the mean
- - - - Medical approach to reducing anxiety is pharmacological - that carries complications
        
        Non pharmacological interventions? Does music work? How can we find out if music works or not?
- - - - Assumes same population
      - Typically used for pure replications
      - In fixed, we can generalise to the same population.
    - - Assumes different populations
      - Typically used for conceptual replications
      - True fixed effect is rare in psych, we'll usually see random effects model.
      - In random effects, we can generalise across populations.
- - - - Variable #1
      - Variable #2
      - Moderation Effect (Variable #1 x variable #2)
    - - Main effect*: How much one of the predictors in your regression is associated with the outcome.
      - Moderation effect or interaction effect*: Represents the combination of predictors, which needs visualisation to be interpreted.
      - Displaying effects:
        
        Main effect of predictor only: A stronger main effect of the predictor makes the angle of both lines steeper (like a correlation).
        
        Main effect of moderator only: A stronger main effect of the moderator makes one line higher than the other.
        
        Moderation effect:
        
        The lines meet exactly in the middle, neither one is higher in total, this is the moderation effect only.
        
        The average angle of each line cancels out
        
        No main effect of predictors or moderators.
        
        The association depends on the moderator.
        
        Identifying main + moderation effects:
        
        Imagine a line in the middle that combines the slopes. If it's positive or negative, not flat, there is probably a main effect of the predictor.
        
        One is higher than the other, there is a main effect of the moderator.
        
        They are at different angles so there is a moderation/interaction effect.
        
        In Jamovi
- - - - Not predictive, just cross-sectional
      - The link between happiness and productivity in the workplace depends on how happy you are
- - - - Operationalise a psychological phenomenon. (use prior theory and research).
      - Identify the behavioural and emotional components of that phenomenon.
      - Example: The demand / withdraw patterns. One partner "nags" the other partner who then withdraws or "shuts down".
        
        "Demand"
        
        Making demands
        
        Voicing complaints
        
        Criticising the partner
        
        Interruption
        
        "Withdraw"
        
        Behavioural withdrawal
        
        Minimising emotion
        
        Hesitant or no response
        
        Off-topic conversation
    - - Provide example videos of behaviours on the coding schedule to the coders ("anchors").
      - Example:
        
        Anchor for low demand/withdraw. 1/7, "I'll do it", nothing else said, it doesn't happen.
        
        Anchor for high demand/withdraw. 7/7, "I'll do it in a bit", more nagging, goes off topic, "leave me alone I have work"
      - Explain the coding schedule in detail. (but not the hypothesis)
    - - An important form of reliability for observational coding is...
        
        inter-rater reliability or observer agreement: The extent to which different observers' judgements are associated.
        
        Tested by obtaining the inter-class correlations for each behavioural code, between all pairs of coders.
        
        Inter-class correlations for withdrawal coding.
        This is the same process as assessing internal reliability for a measurement scale.
        
        Don't drop below 0.6 (Chronbachs alpha)
        
        Coder drift: over time, observer's coding can be inconsistent…
        
        Observers learn different "anchor points: for high and low levels of behaviour after more varied exposure.
        
        Experience increases coding ability.
        
        Boredom decreases coding ability.
        
        How to prevent it: Code in teams of 2-3. discuss and justify ratings after coding. Re-code the initial videos to examine differences over time.
      - Validity: the extent to which your measurement actually measures what you say/think it measures
        
        External validity: does it associate with/predict what it should in the world?
        
        Tests of validity: validation hypothesiis
        
        Null hyp = the observed variables are unrelated to one another
        
        Alt. hyp 1 = 1 demand is negatively associated with withdrawal
        
        Alt hyp. 2 = 2 withdrawal is negatively associated with satisfaction