Statistics / Correlation research

After a research project has been carried out, what are the results?

For quantitative data, the results are a bunch of numbers.

Now what? What do the numbers look like, what do the numbers mean

Statistical analysis allows us to:Summarize the dataRepresent the data in meaningful waysDetermine whether our data is meaningful or

not

Many forms of researchMany forms of dataVariety of dependent variables

Data can take 1 of 4 different forms. Four measurement scales:

NominalOrdinal IntervalRatio

Nominal scale – simplest form of measurement: you give something a name. Qualitative scale of measurement

Assign participants to a category based on a physical or psychological characteristic rather than a numerical score.

E.g.,Male vs. Female; color of eyes Intelligence levels: smart vs. dull

Data is determined by a strict category Only allows for crude comparisons of

results. Can really only be used for qualitative

comparisons.

Ordinal scales ranking system – data is ranked from

highest to lowest Show relative rankings but say nothing

about the extent of the differences between the rankings.

Does not assume that the intervals between rankings are equal.

E.g., rank 10 smartest kids E.g., college football rankings Problem – no absolute magnitude Makes it difficult to make comparisons

Interval scales – numeric scores without absolute zero

Not only relative ranks of scores, but also equal distances or degrees between the scores.

Interval = equal intervals ordering E.g., IQ scores – difference between 100

and 120 is the same as the difference between 60 and 80.

Problem – no absolute zeroCannot have an IQ score of 0.Does not allow for ratio comparisons. E.g., IQ

of 120 is not twice as smart as 60.

Ratio scales - numeric scores but with an absolute zero point

All of the properties of the other scales but with a meaningful zero point.

Allows you to make ratio comparisons i.e., is one twice as much as another?

E.g., number of correct answers on an exam.

E.g., number of friends a person has.

Nominal and Ordinal scales are discrete or categorical

Interval and Ratio scales are continuous scales.

NOIR Increasing levels of resolution Most observable behaviors are

measured on a Ratio scale. Most psychological constructs are

measured on an Interval scale. Important to recognize what scale of

measurement is being used. Nominal and ordinal data require different

statistical analyses than interval or ratio data.

After data collection is finished, the data must be summarized. What does it look like?

Start with exploring the data. Look at individual scores.

Frequency distributions show us the collection of individual scores.

Simple frequency distributions – lists all possible score values and then indicates their frequency.

Allows us to make sense of the individual scores.

Bob 80 Axel 86Jenny 70 Jordan 96Joe 66 Marissa 100George 78 Jackson 86Lori 100 John 78Sherri 88 Janice 76Joey 68 Amy 78Cedric 78 Gene 50Jan 56 Dorothy 76Arthur 86 Patrick 80Ackbar 76 Nicole 72Robert 98

Score Frequency100 298 196 188 186 380 278 476 372 170 168 166 160 156 1

Grouped frequency distribution – raw data are combined into equal sized groups

Grade FrequencyA (90 - 100) 4B (80 - 90) 6C (70 - 80) 9D (60 - 70) 2

F(<60) 2

Histogram – a frequency distribution in graphical formBar graph

Numeric summaries that condense information Numbers that are used to make comparisonsNumbers that portray relationships or

associations. Two main types of stats

Descriptive statistics Inferential statistics

Descriptive statistics – summarize resultsCentral tendencyVariability

Inferential statistics – Used to determine whether relationships or differences between samples are statistically significant

Central tendency – what is the “heart of the data”?

Three measures of central tendency Mean – average

Add up all scores and divide by the total number of samples

Median – middle scoreLine up all scores and find the middle one

Mode – most common scoreWhich score occurs the most often

Simply add up all of the scores and divide by the number in the sample.

The statistic for a sample – X bar - = X / n

X̄ X̄

X̄

Bob 80 Axel 86Jenny 70 Jordan 96Joe 66 Marissa 100George 78 Jackson 86Lori 100 John 78Sherri 88 Janice 76Joey 68 Amy 78Cedric 78 Gene 50Jan 56 Dorothy 76Arthur 86 Patrick 80Ackbar 76 Nicole 72Robert 98

Total 1822n 23

= X / n = 1822 / 23 = 79.22

Pros and cons of using the mean Pros

Summarizes data in a way that is easy to understand.

Uses all the data Used in many statistical applications

Cons Affected by extreme valuesE.g., If Robert would have scored a 0, the

mean changes to 74. E.g., average salary at a company

12,000; 12,000; 12,000; 12,000; 12,000; 12,000; 12,000; 12,000; 12,000; 12,000; 20,000; 390,000

Mean = $44, 167

Median – the middle score in the data: half the scores are above it, half of the scores are below it.

Scores are ranked…. Find the one in middle.

50 56 66 68 70 72 76 76 76 78 78 78 78

80 80 86 86 86 88 96 98 100 100

Example – Median is the score 78. If there is an even number of scores, the

median is the average of the two middle scores.

E.g., 10, 10, 9, 9 – Median is 9.5

Pros and cons of using the median Pros

Not affected by extreme values Always exists Easy to compute

Cons Doesn't use all of the data values Categories must be properly ordered

Mean is almost always preferred. Exception: data is skewed, not distributed symmetically, or has extreme scores.

Positive Skew

27 32 37 42 47 52 57 62 67 72 770

2

4

6

8

10

12

Scores

Freq

uenc

y

Negative Skew

27 32 37 42 47 52 57 62 67 72 770

2

4

6

8

10

12

Scores

Fre

qu

en

cy

Mode – the most common score of the data Mode is 78

Score Frequency100 298 196 188 186 380 278 476 372 170 168 166 160 156 1

Pros and cons of using mode Pros

Fairly easy to computeNot affected by extreme values

Cons Sometimes not very descriptive of the data Not necessarily unique – if two modes =

bimodal; if multiple modes = polymodal.Doesn't use all values.

Examples: shoe size, height

Variability – how spread out is the data Measures of variability

RangeVariance Standard deviation – “average variability”

Range – the simplest variability statistic = high score – low score.

Standard deviation - a measure of the variation, or spread, of individual measurements; a measurement which indicates how far away from the middle the scores are.

The larger the standard deviation, the more spread out the scores are.

The smaller the standard deviation, the closer the scores are to the mean.

Computing SD1. subtract each score from the mean

Ex. (100 – 80 = 20)2. square that number for each score3. add up the squared numbers. This is the

“sum of squares” 4. Divide the sum of squares by the total

number in the sample minus one - this is the variance

4. take the square root of that number. This is the standard deviation

Data is usually spread around the mean in both directionsSome are higher than the mean, some are

lower. The frequency distribution of the scores

tells us how the scores land relative to the mean.

Ideally, some scores are higher, some are lower, most are in the middle.

The normal distribution – the bell curve

As sample size increases, the distribution of the data becomes more normalized.

Importance of the normal distributionSymmetricalMean, median, mode all the sameThe further away from the mean, the less likely

the score is to occurProbabilities can be calculated

We can assume that many human traits or behavior follow the normal distribution

Some are high is a trait, some are low, but most people are in the middle.

E.g., personality traits, memory ability, musical capabilities

People have a tendency to think categorically - erroneous

All data points are arranged, and a particular data point is compared to the population.E.g. IQ score of 130

Percentile reflect the percentage of scores that were below your data point of interest. IQ score of 130 is at the 95th percentile.

Percentile is arranged according to standard deviation.

0 SD is the 50th percentile

1 SD is the 84th percentile

2 SDs is the 97th percentile

3 SDs is the 99.5th percentile

Advanced statistics that reveal whether differences are meaningful.

Take into account both central tendency (usually the mean) and variability

Determines the probability that the differences arose due to chance.

If the probability that the observed differences are due to chance is very low, we say that the difference is statistically significant.

Science holds a strict criteria for determining significance.

α = alpha – the probability of committing a Type I error.

α is normally set at 0.05. Only a 5% chance of committing a type I error.

Can find the probability that the observed differences are statistically significant. If that probability is less than 0.05, the results

are statistically significant. Many types of inferential statistics

t testAnalysis of Variance

Visually representing the data can make it more understandable for you as well as anyone else looking at your results.

Horizontal axis is the X-axis Vertical axis is the Y-axis The best graph is the one that makes the

data more clear.

50.00 60.00 70.00 80.00 90.00 100.00

Scores

0

2

4

6

8

10

Fre

qu

en

cy

Mean = 79.2174Std. Dev. = 12.75987N = 23

Each score is divided into two parts, a stem and a leafThe leaf is the last digit of the scoreThe stem is the remaining digit(s)E.g., 49 would have 4 as the stem and 9 as the

leaf. Graphing a stem and leaf is like making a

table.

Stem Leaf

5 6

6 068

7 0266688888

8 006668

9 68

10 00

Much of the time, a plot of the means is useful.

Test of Men vs. Women

0

10

20

30

40

50

60

70

80

90

100

Female Male

Sco

res

Line graphs are especially important for Repeated Measures

Latent Inhibition

0

10

20

30

40

50

60

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Sessions

Res

pons

es p

er M

inut

e

CS -Preexposed

Control

Show the median and distribution of scores.

Also shows outliers – scores that are more than 3 standard deviations from the mean.4050N =

TRAIT

IntrovertExtravert

FR

IEN

DS

20

10

0

-10

1

Keys to making figures:Keep it simpleNothing is “required” for making figuresPurpose is to better illustrate the results.

Don’t “lie” with figures. Axes should be set at appropriate range.

Test of Men vs. Women

79.4

79.6

79.8

80

80.2

80.4

80.6

80.8

81

81.2

Female Male

Sco

res

Test of Men vs. Women

0

10

20

30

40

50

60

70

80

90

100

Female Male

Sco

res

Correlational research investigates the relationships between two variables.E.g., is there a relationship between poverty

levels and crimeAttachment level in children and future

behavior.Are the number of hours husbands spend

watching sports associated with wives’ marital satisfaction?

Are basketball players heights associated with number of points scored?

Establishes the relationship between the variablesWhether it existsThe strength of the relationship

Correlation can be used as a method for conducting research, or as a tool within the research.

Correlation does not mean causation

Ex. Significant correlation between ice cream sales and murder rates – ice cream sales and shark attacks

The number of cavities in elementary school children and vocabulary size have a strong positive correlation.

Skirt lengths and stock prices are highly correlated (as stock prices go up, skirt lengths get shorter).

Can be causation, but correlational research is not designed to assess that.

Meanings of correlation:1. Causation: Changes in X cause changes in Y2. Common Response: changes in X and Y are

both caused by some unobserved variable.3. Confounding variables are causing Y and not

X.

Correlation simply measure relationships. All methods use to calculate correlation are

established so that it can vary between –1 and +1.

Most common method is the Pearson product-moment correlation coefficient Represented by r

Strength of the correlationThe closer to +1 or -1, stronger the correlation

Positive correlations – as X increases, Y increases.Ex. Horsepower and speedThe value of the correlation represents the

strength of the relationship.+1 represents a perfect positive relationship.0.9 is an extremely high correlation, 0.2 isn’t as

strong. Zero correlations – as X increases, we have no

idea what happens to Y.Values around 0Examples: length of hair and test scores

Negative correlations – as X increases, Y decreases.Horsepower and miles per gallon

Important: a negative correlation simply tells what direction the relationship is, not the strength of the relationship.

One way to view correlations is graphically. Scatterplots – graph that plots pairs of

scores: one variable on the X axis, one on the Y axis.

Concurrent Change Same Direction

Strong positive correlation:

Engine size and Weight

Vehicle Weight (lbs.)

6000500040003000200010000

Eng

ine

Dis

plac

emen

t (cu

. inc

hes)

500

400

300

200

100

0

-100

Very weak correlation

Model year and weight

Vehicle Weight (lbs.)

6000500040003000200010000

Mod

el Y

ear

(mod

ulo

100)

8483828180797877767574737271706968

Strong negative correlation:

Horsepower and miles per gallon

Horsepower

3002001000

Mile

s pe

r G

allo

n

50

40

30

20

10

0

Negative Correlation

Concurrent Change in Opposite Directions

Scatter plots also allow you to see outliers. Most correlations are assessing a linear

relationship. Some relationships are more complex. E.g., the Yerkes-Dodson law