CRICOS Provider No. 00300K (NT/VIC) I 03286A (NSW) | RTO Provider No.

0373

PSY41: Research Methods and

Week 1 – Review of basic statistics

Faculty Of Health
Dr. Rebecca Williams
SEM1 2023
Recognition of Traditional owners
and Indigenous cultures
Charles Darwin University acknowledges the traditional
custodians of the land on which we’re meeting and pays
respect to Elders both past and present and extends that
respect to all Aboriginal and Torres Strait Islander people.

2
Staff Contacts
Dr. Rebecca Williams
• Email: [email protected]
• Phone: +61 0889466829
• Office: Blue 1 Level 1 Room 33
• Campus: Casuarina
• Office hours by appointment

3
4
Unit description

• Let’s go through the course Outline provided on

Learnline

5
SPSS or Jamovi?
• Last year- 3rd year stats with Dr. Rob Heirene- the class
switched to Jamovi as the main statistical analysis
software

• But the prescribed textbook for this course is for SPSS

• Let’s discuss how we should proceed... SPSS or Jamovi

for in-class examples?
6
Today’s learning outcomes
• Understand the basic overall research process

• Know some basics to consider when collecting data

• Refresher on variables

• Recap on some important descriptive statistics

• Frequency and central tendency
• Dispersion
• Probability

• Refresher on inferential statistics (null hypothesis significance testing)

• Remember that it’s not just about the ‘p-value’

7
The research process
• Research is expensive and difficult
• Most Australian research is funded by the government
• One example of costs: from 2022 onwards, the government is providing $650 million each year for
new medical research projects through the Medical Research Future Fund (this is just one of
Australia’s research funding bodies...)
• But it’s worth it: We need an objective method to obtain information, because the human
brain is highly biased. Three cognitive biases are
• Availability heuristic: is when we over-estimate the occurrence of highly emotional or memorable
events
• Confirmation bias: is the tendency to seek and pay attention to the information that supports one’s
belief’s
• Belief perseverance: is holding on to one’s belief, even when faced with disconfirming evidence

8
 The research process: Develop a research
question
A research question can originate
from observation
• But should form after a critical
evaluation of the literature

Despite what this figure from the

textbook says, the reality is that
we do not collect data until we
have a sound research question
and research design
9
Field (2018). Discovering Statistics (5th ed).
 The research process: Generate a hypothesis
A hypothesis is a clear, explanatory statement that can
(although not always) be informed by a theory

The attributes of a good scientific theory:

1. Has plenty of independent support
2. Can be falsified
• Meaning that it is precise enough to be refuted
• This is often the difference between science and
pseudoscience
3. Abides by the law of parsimony
4. Allows for predictions

• A prediction should logically follow a hypothesis

10
Field (2018). Discovering Statistics (5th ed).
 The research process: Collecting data
• Now we look to what we need to do to
collect data

• Of course, we need to get ethics approval

• Variables are things that can vary

Ethics • Independent variable (IV) = the predictor, or
approval the experimental manipulation

• Dependent variable (DV) = the criterion, or

the outcome measurement

11
Field (2018). Discovering Statistics (5th ed).
A quick refresher on variables: Measurement
levels
• Nominal variable = groups (more than 2; binary = 2)
... what about a
• Sex, gender, nationality Categorical Likert scale...??
• Ordinal variable = levels variables
• Ranked preferences
• Interval variable = a score with no zero (+,-) Continuous
• Temperature (in C or F), IQ
variables
• Ratio variable = a score with absolute zero (+,-,x,÷)
• Length, reaction times
12
 The research process: Collecting data
When measuring variables to address a research question, there are some things
to consider:

1. Am I doing correlational or experimental research?

• Or is it quasi-experimental?
• This is where groups are designated based on existing factors so that people cannot
be randomly assigned to groups. For example, sex or nationality

2. Am I doing a within or between-subjects design?

• Or perhaps mixed?
• This is where there are both within and between-subjects factors
Ethics
approval
3. Is the instrument I’m going to use to measure my variable valid and
reliable?
• Criterion (concurrent + predictive) and content validity

4. Am I doing everything I can do reduce unsystematic variation?

• Do I need to do randomisation to reduce practice or boredom effects
(within-subjects), or for group allocation (between-subjects)?

13
Field (2018). Discovering Statistics (5th ed).
 The research process: Analysing data
There is a lot to think about when
analysing data! Let’s consider some
fundamentals:

1. How often does each score occur?

Ethics
approval
2. What is the dispersion of your
scores?

3. What is the probability of obtaining

a score?
14
Field (2018). Discovering Statistics (5th ed).
Menti time
• Menti is a fun, anonymous and non-graded online quiz that we do
together as a group

• Go to this website:
menti.com
And enter this code:

3330 1923
15
 The research process: Analysing data
There is a lot to think about when
analysing data! Let’s consider some
fundamentals:

1. How often does each score occur?

Ethics
approval
2. What is the dispersion of your
scores?

3. What is the probability of obtaining

a score?
16
Field (2018). Discovering Statistics (5th ed).
1. How often does each score occur?
• This is asking for the frequency of each
score

• We need to check this by plotting the

scores in a histogram

• Here are some example IQ scores

from 100 people

• This allows us to do a quick eyeball

check for outliers and normality (but
more on that next week...)
17
1. How often does each score occur?
Measures of central tendency
(i.e. the centre of the
histogram) summarize the data
• Mode = most frequently Σ 2, 2, 4, 6, 7, 8, 13
occurring score
7
• Median = mid score (when
ranked)
• Mean = the average
18
 The research process: Analysing data
There is a lot to think about when
analysing data! Let’s consider some
fundamentals:

1. How often does each score occur?

Ethics
approval
2. What is the dispersion of your
scores?

3. What is the probability of obtaining

a score?
19
Field (2018). Discovering Statistics (5th ed).
 2. What is the dispersion (i.e spread) of the
scores?
• How far is each score from the mean?

• If we calculate the difference between each score

and the mean- and then sum them all- we have
the total deviance

• BUT the deviance will be zero. So, we square each

value before summing. This is the sum of squared
errors (SSE)
......
• But SSE is not a good way to compare between 65 - 99.45 140 - 99.45
datasets, because it depends on the sample size = -34.45 = 40.55
(65 - 99.45)2 (140 - 99.45)2
• Divide this number by the degrees of freedom (N- = 1186.8 = 1644.3
1) to get an average of the SSE: this is known as
the variance (𝑠2) 99 20
Is variance the final number we need to
know the spread of the scores?
• No

• Because we squared the scores (to get the SSE),

𝑠2 is no longer in the same units as our original
measure

• In other words, we want to know the spread of IQ

scores, not the spread of IQ scores squared

𝑡𝑜𝑡𝑎𝑙 𝑑𝑒𝑣𝑖𝑎𝑛𝑐𝑒 = ∑ 𝐼𝑄𝑆𝑐𝑜𝑟𝑒! − 𝑀𝑒𝑎𝑛 𝐼𝑄 𝑆𝑐𝑜𝑟𝑒

• So now we take the square root of the 𝑠2 , which
puts us back into regular IQ scores 𝑆𝑆𝐸 = ∑ (𝐼𝑄! − 𝑀𝑒𝑎𝑛 𝐼𝑄 𝑆𝑐𝑜𝑟𝑒)2
∑ (𝐼𝑄! − 𝑀𝑒𝑎𝑛 𝐼𝑄 𝑆𝑐𝑜𝑟𝑒)
𝑠2 =
• This is the standard deviation (𝑠), which tells us 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐼𝑄 𝑠𝑐𝑜𝑟𝑒𝑠 − 1
the average difference between people’s IQ
scores and the mean IQ score 𝑠=
∑ ($%! & '()* $% +,-.()
21
0123(. -4 $% 5,-.(5 &6
 The research process: Analysing data
There is a lot to think about when
analysing data! Let’s consider some
fundamentals:

1. How often does each score occur?

Ethics
approval
2. What is the dispersion of your
scores?

3. What is the probability of obtaining

a score?
22
Field (2018). Discovering Statistics (5th ed).
3. What is the probability of obtaining a
score?
In statistics we don’t simply
want to know how many
people had a certain IQ
score, but rather what is the
probability of having a P(𝐼𝑄 𝑠𝑐𝑜𝑟𝑒 𝑜𝑓 80) =
!"#. %& '(%')( *+,- ./ %& 01
,%,2) !"#. %& '(%')(
certain IQ score? 31
P(𝐼𝑄 𝑠𝑐𝑜𝑟𝑒 𝑜𝑓 80) = = 0.1 = 10%
311

23
3. What is the probability of obtaining a
score?

Probability (%)
We can now replace
‘frequency’ on the y-axis
with ‘probability’

!"#. %& '(%')( *+,- ./ %& 01

P(𝐼𝑄 𝑠𝑐𝑜𝑟𝑒 𝑜𝑓 80) =
,%,2) !"#. %& '(%')(
31
P(𝐼𝑄 𝑠𝑐𝑜𝑟𝑒 𝑜𝑓 80) = = 0.1 = 10%
311

24
3. What is the probability of obtaining a
score?

Probability (%)
• Now imagine that someone could
score any value IQ score, rather
than an integer
• Example: 100.0021

• We now need to break down our

x-axis into infinitely thinner
rectangles

• When the rectangles become

infinitely small, the histogram
becomes a curve 25
3. What is the probability of obtaining a
score?

Probability (%)
• This curve is known as a
probability distribution

• Instead of asking how many

people had an IQ score of 80, we
can, for example, ask the
probability of obtaining a score
less than 80

• The probability is calculated as the

area under the curve
26
Z-scores

Probability (%)
• So, you’re telling me that I have to calculate the area
under the curve????

• No, that sounds like calculus rather than stats

• We can use a standard probability distribution, which has

already done the hard work for us
-1 0 1

• To use this, we convert our scores to z-scores (mean = 0,

standard deviation = 1)
𝐼𝑄𝑠𝑐𝑜𝑟𝑒+ − 𝑀𝑒𝑎𝑛 𝐼𝑄 𝑆𝑐𝑜𝑟𝑒
• A z-score tells us how many standard deviations our score 𝑧=
is from the mean (0) 𝑠

• The middle 95% of scores fall within z = -1.96 and 1.96

27
Let us now transition from descriptive to
inferential statistics...
• So far we have described our data - IQ scores - in terms of
• Frequency and central tendency (mode, median, mean)
• Dispersion (variance, standard deviation)
• Probability of obtaining a single (or range of) IQ scores (z-scores)

• But we also like to run statistical tests to obtain information about

the population rather than just our sample

• Which we do by taking our sample mean (rather than individual

scores) and compare it to... the null hypothesis!
28
TIME FOR BREAK?

29
Null hypothesis significance testing (NHST)

• When we run statistical tests (t-test, ANOVA, etc.), this test

will always ask one simple question:

What is the probability of obtaining our results if our data

were purely random?

• This is known as the the null hypothesis (H0)

30
We test our data against the null hypothesis
• This is when we calculate the probability of obtaining
our data if they were random
• i.e. Are our data different from the null hypothesis (H0)?

• There are only 2 possible outcomes:

1. No, our data are not different from H0: Fail to reject H0
2. Yes, our data are significantly different from H0: Reject
H0
31
We can do this for our IQ scores
• Assume that we have IQ scores from group 1, and a
second set of scores from group 2. Group 1 Group 2
99 96
100 100
• We want to know whether there is a difference 101
101
100
102
between the means of the groups 102 104
103 106
104 109
113 115
• H0 = there is no difference between the means of group 120 121
1 and group 2 132 127

• Alternative hypothesis (H1) = there is a significant

difference between group 1 and group 2 means 32
What do I mean by ‘significant’
• This means that we can reject H0 with a high
degree of confidence Group 1
99
Group 2
96
100 100
101 100
101 102
• We have to set this confidence by creating a cut-off 102
103
104
106
or a threshold 104
113
109
115
• “What is the probability of incorrectly rejecting the 120
132
121
127
null hypothesis?”
• Less than 5% (0.05) is standard minimum threshold
• This threshold also referred to as the alpha (𝛼)
33
Let’s go back to our IQ scores...
• Assume that we have IQ scores from group 1, and a second
set of scores from group 2. Group 1 Group 2
99 96
100 100
• We want to know whether there is a difference between the 101 100
means of the groups 101 102
102 104
103 106
104 109
• H0 = there is no difference between the means of group 1 113 115
and group 2 120 121
132 127

• Alternative hypothesis (H1) = there is less than 5% probability

of getting these scores if there was no difference between
group 1 and group 2 means 34
H1 can be directional or non-directional
• Our H1 is an example of a non-directional test :

“Alternative hypothesis (H1) = there is less than 5%

probability of getting these scores if there was no
difference between group 1 and group 2 means”

• This means it’s a two-tailed test: group 1 or group 2

could have the higher mean score

• A one-tailed test would be directional

• Example: Group 1 would have significantly higher
scores than group 2
35
We run a statistical test, we get a test
statistic (e.g. t-score) and it’s p-value
• Which stands for ‘probability’

• This is the probability of obtaining our test statistic if the null hypothesis is true

• We compare our calculated p-value to alpha (threshold)

• If p < alpha, our result is significant (i.e. we can reject the null hypothesis)
You can find a
• So, what’s the difference between p-value and alpha? flowchart of the
• We set alpha as the cut-off NHST process on
• P-values are calculated from the data (we don’t set these) page 76 of your
textbook 36
Why is alpha set to 0.05?
• It doesn’t have to be

• But it comes from the standard normal distribution we just spoke about
• 95% of data fall within 2 standard deviations of the mean
• We want our results to be distinct, i.e. in the tail-ends of the null distribution

37
 What are the caveats of setting alpha too
high or too low?
Type I and Type II errors Is there really a significant effect?

Does our test indicate that there is a

Yes No
• Type I error is a false positive
• We think there is a significant result when there Correct

significant effect?

Yes
isn’t one Rejection of H0 Type I Error

• Type II error is a false negative

• We think there is no significant result when where Correct failure to

No
really is one Type II Error reject H0

• We can try to avoid Type I errors by decreasing

alpha….

• But this increases the possibility of a Type II error 38

Methods to reduce chances of Type I error
• Remember that the familywise error rate (likelihood of Type I error)
increases with number of tests performed

• Use ANOVA instead of multiple t-tests

• Plan post-hoc tests carefully

• Use Tukey’s HSD

• Use an alpha correction technique (e.g. Bonferroni) when running

multiple tests
39
Does the p-value tell me everything about a
statistical test?

• It’s important, but there are other values you must

check/report when reading and writing a scientific
paper

• These include the

• Confidence intervals (CI)
• Effect sizes
40
 Confidence intervals (CIs) remind us that our
data are noisy
CIs are typically set to 95%, which means

" The population mean lies within this range

in 95% of samples”

• Lower and upper boundaries are on

either side of the mean

• A more narrow range = what we want

(less noise, larger effect size)
41
Field (2018). Discovering Statistics (5th ed).
Effect sizes remind us that p-values are linked
to sample size
... and that the magnitude of the effect isn’t given with the p-value.

• Effect sizes allow for comparisons across studies

• Cohen’s d is the most common method to calculate effect size
• It is the difference between group means divided by the standard
deviation:
𝑋4 1 − 𝑋4 2
𝑑=
𝑠
• s can be chosen from one group, or it can be pooled across both
groups

• d = 0.2 (small), d = 0.5 (medium), d = 0.8 (large) 42

Today’s learning outcomes
• Understand the basic overall research process
We looked at the overall process via a flow chart (initial observation to data analysis)
• Know some basics to consider when collecting data
• Refresher on variables
Nominal, ordinal, interval and ratio variables
• Recap on some important descriptive statistics
• Frequency and central tendency
• Dispersion
• Probability
Mode, median, mean; variance and standard deviation; frequency to probability and z-scores
• Refresher on inferential statistics (null hypothesis significance testing)
H0 and H1; alpha and p-values; Type I and II errors
• Remember that it’s not just about the ‘p-value’
Confidence intervals and effect sizes 43
Other important concepts to review
• How to report data (end of Chapter 1)

• Parameter estimates (Chapter 2)

• Statistical power (Chapter 2)

44
Menti recap quiz
• Menti is a fun, anonymous and non-graded online quiz that we do
together as a group

• Go to this website:
menti.com
And enter this code:

8563 6213
45