KHA350 Lecture Notes - Lecture 8: Lincoln Near-Earth Asteroid Research, Polynomial Regression, Simple Linear Regression
25 Jun 2018
School
Department
Course
Professor
Research methods week 8: regression analysis and analysis of trends
- Not looking at the differences between means but the RELATIONSHIPS
- Use these to build models of the relationships
oTo predict people’s scores on one variable
oTo understand the magnitude of the relationship and the number of
things that may influence other variables
- Regression models:
oLess powerful
oNeed greater sample size in order to achieve significance
oAnswer different questions to something like ANOVA
oComplimentary to each other:
Does mindfulness therapy improve peoples depression
Test for significant decline in depression: tells us whether the
technique worked or not
Regression: tells you things that are important; why something
worked
Picking the right tool for the job:
- so far we have discussed:
ot-tests or one way ANOVA: 1 IV, 2 levels, 1 DV
Independent samples if between subjects
Matched paired repeated measures in within subjects
oOne way ANOVA: 1IV, 2 or more levels, 1 DV
Between or within subjects
oFactorial ANOVA: 2 or more IV, 2 or more levels in each, 1DV
Completely between, within or mixed
Allows additional analysis of whether the IVs independently or
additively impact on the DV
- Idea: build a flow chart of each of these
oHow many iv’s
oWhat the measures are
oAnd how you will analyze each
- These techniques all have had:
oContinuous DV:
Interval: ordered, constant scale, but no natural zero
Date and temperatures
Ratio: ordered, constant scale, with natural zero
Height, length, age
oCategorical IV:
Nominal Scale: male or female etc. classifications
Ordinal scale: ordered classifications: the intervals may not be
consistent
- Previous techniques ask whether there are differences in the DV due to
different IV/levels
- Some research questions require you to look for relationships between
variables rather than differences between means
oEg. Does MDMA use cause neuropsychological impairments?
oCould examine this in a between groups design
find more resources at oneclass.com
find more resources at oneclass.com
IV: control vs. users
DV: neuro test performance
oIf it does cause damage, is it more informative to look for a
relationship between how much people use MDMA and their cognitive
performance
Relationship between two continuous variables
This is a correlational design
- Correlation revision:
oQuantifies the relationship between two variables
oCorrelation coefficient values ® vary between -1 and +1
oYou get two pieces of information about the relationship from the
correlation coefficient:
The direction:
Positive relationship if one variable increased when the
other increase
Negative relationship if one variable decreases when the
other increases
The magnitude:
Weak relationship is close to zero
Strong relationship is closer to +1 or -1
oMore clustered together on a scatterplot
oThe closer to 1 the more tightly clustered the
relationship is around a line
oThe smaller the value the more diffuse
oScatterplots
- Regression: extends correlation information
oCan build a model
also referred to as modeling relationships
onot just how they relate to one another, but can predict scores from the
model
Predict one variable from knowing the other
oBuild a trend line:
Mathematical equation
Line of best fit: between the variables (IV and DV)
Can use equation to predict scores
oThe equation of a straight line:
Only need to know:
Y intercept value: the value that scores on our
measurement when the IV is zero
The slope of this line: how much of an increase there is
per IV increase
oOne unit increase
Y = bX + a
Y = the predicted value of Y
X = the value of the predictor
a = the intercept on the Y axis
B = the slope of the regression line
find more resources at oneclass.com
find more resources at oneclass.com
With this equation we can predict Y for any value of X
Within the range of values used in model development
oRegression equation is chosen to minimize the differences between
each data point and the line
oSums of the squared difference between the actual and predicted values
of Y
oIs knowing this line actually useful at predicting people’s scores
Could you just use an average of everyone’s scores to predict
And is the slope term any different to zero
- Example: relationship between stressors and mental health symptoms you
have
oScatterplot of the DV
oLooks like, the more daily stressors the more severe the symptoms
Even with no stress there are still some symptoms
oSPSS set up:
2 continuous variables: one column or each of these variables
in datasheet view
each row relates to one individual
scale variables: THIS IS IMPORTANT
Terminology:
IV and DV
Outcome/response variables: DV
oMeasuring the response or outcome
Predictor variables: IV
oUsing these variables to do the predicting with
ANALYSE – REGRESSION – CURVE ESTIMATION
Model options: different ways of describing the variable
relationship
oChose linear: straight line relationship
Ask for ANOVA table
Plot models
Output:
Model summary: overall summary of how good the
model is at predicting DV
oHow good the overall model is
oR Value: correlation between the variables
Positive/negative
Magnitude
As stress goes up so does the number of
mental health symptoms
R ^2 and adjusted: adjusted takes into
account sample size you have
If both present always chose
adjusted
Tells magnitude of relationship
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Research methods week 8: regression analysis and analysis of trends. Not looking at the differences between means but the relationships. Use these to build models of the relationships: to predict people"s scores on one variable, to understand the magnitude of the relationship and the number of things that may influence other variables. Regression models: less powerful, need greater sample size in order to achieve significance, answer different questions to something like anova, complimentary to each other: Test for significant decline in depression: tells us whether the technique worked or not. Regression: tells you things that are important; why something worked. Picking the right tool for the job: so far we have discussed: t-tests or one way anova: 1 iv, 2 levels, 1 dv. Matched paired repeated measures in within subjects: one way anova: 1iv, 2 or more levels, 1 dv. Between or within subjects: factorial anova: 2 or more iv, 2 or more levels in each, 1dv.
