Reporting Linear Regression

Question 1

MODEL CHECKING

Answer 1

• to know if regression model is good/not via residual values, use:
  1. general residuals
  2. unusual observations/outliers
  3. residual plots in SPSS

Question 2

GENERAL RESIDUALS

Answer 2

• predicted value for y (DV)
• actual (observed) value for y
• residual value = actual MINUS predicted
• good best fit = small residuals
• moderate fit = larger residuals

Question 3

RESIDUALS: DV vs IV

Answer 3

• difs between actual/predicted values (ie. residual values) should show normal distribution
• some large positive/negative BUT mostly small = normal distribution

Question 4

RESIDUALS: CRITERIA

Answer 4

• should be:
  1. normally distributed (some large negative/positive BUT most small/0)
  2. independent (no constant covariation)
  3. almost identical in variance terms (regardless of IV/DV values)
• easy to check w/SPSS plots

Question 5

UNUSUAL OBSERVATIONS = OUTLIERS (?)

Answer 5

• data may contain cases for which model simply doesn’t work (ie. large difs between observed value of DV vs predicted by model; reflected as large residual value)
• such cases may be:
  1. extreme scores within normal range
  2. members of dif sub-group entirely aka. outliers
• but how to tell which?

Question 6

DISTRIBUTION OF RESIDUAL VALUES

Answer 6

HISTOGRAM
- for residuals w/normal curve
FREQUENCY DISTRIBUTION
- for residuals