PY 701 Week 6
Review of How to Write Various
Results
• Always include the statistical information to support your
statements
• Generally, whether the results are significant or not, you
will write p=.##. BUT… if the p value reported on the
output is .000, then you will write p<.001. See APA
manual p. 204
• For descriptive data:
• Continuous variables - report range, mean, and standard deviation
• Categorical variables - report frequency and percent
Correlation
• In order to test the hypothesis that there would be a relationship
between depression and self-esteem, a correlation was run.
Results were significant (r=##, p=.##). Results indicated
that… (e.g., more depressed individuals tend to have lower self-
esteem).
• OR: In order to test the hypothesis that there would be a
relationship between depression and self-esteem, a correlation
was run. Results were not significant (r=##, p=.##).
Independent-Samples t-test
• In order to test the hypothesis that there would be a difference between the treatment
group and the control group on depression levels, an independent samples t-test was
run. Results were significant (t(df)=##, p=.##). The average depression score in the
treatment group (M=##, SD=##) was significantly lower than the average depression
score in the control group (M=##, SD=##).
• OR: In order to test the hypothesis that there would be a difference between the
treatment group and the control group on depression levels, an independent samples t-
test was run. Results were not significant (t(df)=##, p=##).
• *Note: df = degrees of freedom. When you report your results, don’t write “df” – write
in the number of degrees of freedom for that analysis. You can find that number on the
SPSS output. Also, replace the ## signs with the values for these things from the SPSS
output.. Most (but not all) statistical symbols should be in italics (see APA manual p.
187).
ANOVA
• In order to test the hypothesis that there would be a difference between the
treatment group, the placebo group, and the control group on depression levels,
an ANOVA was run. Results were significant (F(dfbt, dferror)=##, p=.##).
Follow-up post-hoc* indicated that there was a significant difference
between medication* and combined treatment(Mdiff=12.68, p=.01) and a
significant difference between behavioral tx and the combined
tx(Mdiff=11.06, p<.001). There was not a significant difference between
the medication tx and behavioral tx(Mdiff=1.62, p=.87).
• OR: In order to test the hypothesis that there would be a difference between the
treatment group, the placebo group, and the control group on depression levels,
an ANOVA was run. Results were not significant (F(dfbt, dferror)=##, p=##).
Simple Linear Regression
• In order to test the hypothesis that level of self-esteem would
predict level of depression, a simple linear regression was done.
Results were significant (F(regression, residual)=##, p=.##).
The overall model accounted for ##% of the variability in
depression (β=##, p =.##).
• OR In order to test the hypothesis that level of self-esteem would
predict level of depression, a simple linear regression was done.
Results were not significant (F(regression, residual)=##, p=.##).
Hierarchical Linear Regression
• Hierarchical Linear Regression: In order to test whether a combination of
self-esteem and severity of trauma history would predict depression better
than self-esteem alone, a multiple regression was run. The overall model was
significant (F(regression, residual)=##, p=.##) and accounted for ##% of
the variability in depression. The model using both self-esteem and trauma
history predicted depression significantly better than the model with self-
esteem as the only predictor (R2 change = ##, p=.##). Self-esteem
contributed significantly to the model (β=##, p=.##). Trauma history also
contributed significantly to the model (β=##, p=.##).
• *note: instead of R2 change = ##, p=.##, could put Fchange(df)=##, p=.##
• *note: β = standardized beta weight
• Those are just a few examples. Follow that same general
idea for all reporting of statistical information.
• In preliminary analyses and exploratory analyses, don’t
say “in order the test the hypothesis” because you are not
testing hypotheses.
• Ex: to determine whether…
• Ex: to further explore…
Statistical Decision Tree
• [Link]
re%2012_EN%20Workshop%20Field%202011%20Decision%
[Link]
• Or: Andy Field Textbook p. 776 (slightly different from link
above)
• On decision tree:
• Outcome variable=dependent variable
• Predictor variable = independent variable
• Practice reading decision tree generally then with following
statistics
Evaluating
Relationships/Predictions
• Pearson Correlation or Simple Linear Regression: assess degree of relationship/prediction
between 2 continuous variables. (If prediction- regression)
• Ex: GPA, SAT
• Multiple Regression: assess the degree to which 1 continuous variable can be predicted
from a combination of several continuous predictors. (1 continuous DV; 2+ continuous IVs)
• Ex: predict SAT(DV) from GPA and IQ(IV)
• Hierarchical Regression: assess the degree to which adding continuous predictors
increases ability to predict one DV. (1 continuous DV; 2+ continuous IVs)
• Ex: can we predict SAT better with a combo of GPA & IQ than with GPA alone
• Chi Square/Loglinear Analysis – assess relationship between 2 categorical variables (If
prediction – loglinear analysis)
• Ex: Grade in math (e.g., A, B, C, D, F) and SAT classification (meets college readiness benchmark:
yes/no)
Evaluating Differences
• Independent Samples T-test: 1 continuous DV (outcome); 1 categorical
IV (predictor) w/ 2 categories; different people in the 2 categories
• Ex: is GPA different for those who meet SAT college readiness benchmark and
those who don’t
• Paired/Dependent Samples T-test: 1 continuous DV (outcome); 1
categorical IV (predictor) w/ 2 categories; same people in the 2
categories
• Is SAT score different before and after students take a college readiness course
Evaluating Differences
• One-way ANOVA: 1 continuous DV (outcome); 1 categorical IV
(predictor) w/ >2 categories; different people in the categories
• Is GPA different in those who exceed college readiness benchmark,
meet college readiness benchmark, and don’t meet college readiness
benchmark
• Repeated Measures ANOVA: 1 continuous DV (outcome); 1
categorical IV (predictor) w/ >2 categories; same people in the
categories
• Is SAT score different before, immediately after, and 6 months after
taking college readiness course
Evaluating Differences
• One-way ANCOVA: 1 continuous DV (outcome); 1 categorical IV (predictor)
w/ 2+ categories & a covariate; different people in the categories
• Is GPA different in those who exceed college readiness benchmark, meet college
readiness benchmark, and don’t meet college readiness benchmark when IQ is
controlled for
• Factorial ANOVA – 1 continuous DV; >1 categorical IV
• Does GPA differ across levels of college readiness benchmark categories (exceeds,
meets, does not meet) and whether person has had a college readiness course
(Yes/No)
• Factorial ANCOVA – 1 continuous DV; >1 categorical IV, plus covariate
• Same as above, control for IQ
Analysis of covariance
• This is used when you want to compare group means but
you also want to adjust for (or control for) another
variable (or variables) that are expected to be related to
the DV.
• Covariate – the variable that you want to control/adjust for
• Including a covariate can help reduce error (or
unaccounted for variance in the DV)
• *remember fundamental and powerful concept –
explaining variance
Example
• You want to evaluate the effect of alcohol on fine motor
skills. You randomly assign people to 4 groups (3
nonalcoholic beers, 3 beers w/ alcohol, 6 nonalcoholic
beers, 6 beers w/ alcohol). Then you test fine motor skills.
• IV? Levels of IV?
• Alcohol; 4 levels
• DV?
• Fine motor skills
Other variations:
• Factorial ANCOVA
• Add covariate to factorial ANOVA
• Factorial MANCOVA – multiple IVs, multiple DVs, and a
covariate
MANOVA
• Multivariate analysis of variance (MANOVA) is a multivariate
extension of analysis of variance
• Multiple DVs
• MANOVA evaluates whether the population means on a set
of dependent variables vary across levels of a factor or
factors
When to run a Manova and why
(Field, 2018)
• To test for differences among groups
• When we have Several Outcome variables (DVs)
• Better than Multiple ANOVA
• Controls familywise error rate (Type I error)
• Takes account of relationships between DVs
Assumptions of MANOVA
• Assumption 1: The dependent variables are multivariately
normally distributed for each population, with the different
populations being defined by the levels of the factor
• Assumption 2: The population variances and covariances
among the dependent variables are the same across all levels
of the factor. Assumption of homogeneity of the variance-
covariance matrices assessed with Box’s M statistic.
• Assumption 3: The participants are randomly sampled
MANOVA SPSS
Statistics to evaluate the MANOVA hypothesis,
• Wilks’ Lambda
• Pillai’s Trace
• Hotelling’s Trace
• Roy’s Largest Root
MANCOVA
• Removes the effect of covariates (how IVs act on DVs)
• Requires large sample size
• IV – Categorical, DV- Continuous, CV – categorical/continuous
• Assumptions
• Independence of observations
• Normal distribution
• Homogeneity of covariance
• Homogeneity of variance
• Similar regression line slope
• Correlation between DV and CV
Run Manova & Mancova
• Analyze --->GLM-->Multivariate--> DVs in dependent
box--> IV in Fixed factors--->Contrast--->Simple-->EM
means-->Options
Multiple DVs
• One-way MANOVA – multiple continuous DVs, 1 categorical IV
• Does academic readiness for college (as measured by GPA & academic self-efficacy) differ
across those who have taken a college readiness course and those who have not?
• One-way MANCOVA – multiple continuous DVs, 1 categorical IV, covariate
• Same as above, control for IQ
• Factorial MANOVA – multiple continuous DVs, multiple categorical IVs
• Does academic readiness for college (as measured by GPA & academic self-efficacy) differ
across levels of college readiness benchmark categories (exceeds, meets, does not meet) and
whether person has had a college readiness course (Yes/No)
• Factorial MANCOVA - multiple continuous DVs, multiple categorical IVs, covariate
• Same as above, control for IQ
Predicting Group Membership
• Loglinear Analysis - predict group membership (categorical DV) when
predictors are categorical
• Predict level of college readiness benchmark categories (exceeds, meets, does
not meet) from whether a student has had an algebra class (yes/no) and a
writing class (yes/no)
• Logistic Regression – predict group membership (categorical DV)
when predictors are continuous, categorical, or a combination of both
• Predict level of college readiness benchmark categories (exceeds, meets, does
not meet) from GPA
• Or: Predict level of college readiness benchmark categories from GPA,
academic self-efficacy, whether a student has had an algebra class (yes/no), and
whether a student has had a writing class (yes/no)
MCQs
• What test would you run if you were trying to see how breed of
cats (Persian Cat, Sphynx, and Siamese) differ on measures of
feline athleticism.
• One-way Manova
• Factorial ANOVA
• One-way anova
• logistic regression
MCQs
• You are doing a study looking at creativity in adults with ADHD. You have
three participant groups (ADHD (medicated), ADHD (non-medicated), and
non-ADHD) and three dependent variables relating to domains of creativity
including divergent thinking, convergent thinking, and creative achievement.
You would like to analyze the differences between participant groups for
each dependent variable. Which analysis would be most appropriate to run?
• Three-way ANOVA
• MANOVA
• Factorial ANOVA
• Independent T-Test
MCQs
• You wish to conduct a study examining if education level has an effect on anxiety
level and IQ scores in adults, but want to control the effects of age. If education
level is measured based on the degree a participant has earned (High school
diploma, Associate's, Bachelors, Masters, Doctorate) and both anxiety and IQ are
scaled measures, what would be the appropriate statistical test for this study?
•
• a) One-way MANOVA
• b) One-way MANCOVA
• c) Regression
• d) One-way ANCOVA
Develop examples/multiple choice test questions for the
following designs:
• One-way ANOVA • One-way MANOVA
• Repeated Measures • One-way MANCOVA
ANOVA • Factorial MANOVA
• One-way ANCOVA • Factorial MANCOVA
• Factorial ANOVA • Logistic regression
• Factorial ANCOVA
• Loglinear Analysis