deviations from normalitynoir r. atio • satisfies all the requirements of an interval scale •...

52
Deviations from Normality Skewness and Kurtosis

Upload: others

Post on 27-Mar-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Deviations from NormalitySkewness and Kurtosis

Page 2: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Why do we care if the distribution is not normal?

• It helps you understand how a characteristic exhibits itself in your sample or in the population.

• It impacts what descriptive statistics you might use.

• It impacts the inferential statistics you might use.

Page 3: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

SkewnessThe majority of scores do not fall in the middle of

the distribution.

The distribution is asymmetrical

You label the kind of skew according to the longer tail of the distribution.

Page 4: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Normal versus Skewed Distributionsen.wikipedia.org/wiki/Image:Standard_deviation_diagram.png

Freq

uenc

y

Negative skewPositive skew

Long tail is on the negative endLong tail is on

the positive end

Page 5: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Mean

Median

Mode

Normal Distribution

Page 6: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

What Positive Skew MeansIf this represented the results of a quiz, the majority of the participants scored very low— almost no one scored in the highest range.

12

34

5

0 10 20 30 40 50 60 70 80 90 100

This test must have been very

easy or difficult?

Median

Mode

Mean

Page 7: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

What Negative Skew MeansIf this represented the results of a quiz, the majority of the participants scored high— almost no one scored in the lowest range.

This test must have been very

easy or difficult?

0 10 20 30 40 50 60 70 80 90 100

12

34

5

Mean

Median

Mode

Page 8: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Normal vs. Skewed Distributions

Page 9: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Income Distributionshttp://www.city-data.com/city/Chicago-Illinois.html

Page 10: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Kurtosis

leptokurticplatykurtic

Page 11: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

KurtosisLeaping

leptokurticdistribution

0 10 20 30 40 50 60 70 80 90 100

Page 12: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Platykurtic

0 10 20 30 40 50 60 70 80 90 100

Platykurtic like a

platypus distribution

Page 13: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

leptokurtic

Normal (mesokurtic)

platykurtic

Page 14: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Sprinthall - Quick Kurtosis RuleWhen you have a distribution and know its standard deviation and range, you can estimate its kurtosis.

Fact: For a normal distribution, the standard deviation is about 1/6 of the range.

If the standard deviation is more than 1/6 of the range, then a distribution is platykurtic.

If the standard deviation is less than 1/6 of the range, then a distribution is leptokurtic.

How you determine this:Range = Standard Value

6Compare actual sd to SV

If sd > SV, platykurtic

If sd < SV, leptokurtic

Page 15: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Dr. Bellini’s MCC Research

47/6 = SV

SV = 7.83

8.98 > 7.83

Platykurtic

Page 16: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Dr. Bellini’s MCC Research

19/6 = SV

SV = 3.17

3.52 > 3.17

Platykurtic

Page 17: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Skew

$20,000 $40,000 $60,000

Beginning Salary

25

50

75

100

125

Cou

nt

$25,000 $50,000 $75,000 $100,000 $125,000

Current Salary

0

25

50

75

100

Cou

nt

Page 18: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

C:\Program Files\SPSS\University of Florida graduate salaries.sav

10000 20000 30000 40000 50000 60000

Starting Salary

25

50

75

100

125

Cou

nt

Statistics

Starting Salary1100

26064.2026000.00

200006967.982

58300

NMeanMedianModeStd. DeviationRange

Page 19: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Measurement Scales

Page 20: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Measurement

• Assigning numbers to observations following a set of rules.

Page 21: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

How are numbers assigned to observations?

What scale is used?1. Nominal2. Ordinal3. Interval4. Ratio

Page 22: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Nominal Data• Using numbers to label categories, but the

numbers have no inherent numerical qualities• Male = 1 Female = 2

• social security number • jersey numbers• race/ethnicity

Page 23: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Other examples of nominal scaling• Whether a participant does or does not have a

driver’s license (0,1)

• Whether the participant belongs to the experimental group or the control group (0,1)

• The school the participant attends (1,2,3,4,5)

Page 24: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Uses of nominal data• Generally, the most you can do with nominal data

is count it.

Page 25: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Categorical or Continuous?• Is a variable that uses a nominal scale of

measurement categorical or continuous?

Page 26: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Ordinal Scaling• The assigned number provides information about

the rank of an observation.

• Ordinal scales put observations in order.

• Rating scales are often considered ordinal in how they measure characteristics.

Page 27: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Example

Strongly Agree Neither agree Disagree Strongly agree nor disagree disagree

Statistical 1 2 3 4 5Thinking is my favorite Xclass

Hospital pain scales (1-10)

Page 28: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Ordinal Scales have two basic rules:1. Equality/non-equality rule

2. Greater-than-or-less-than rule

Page 29: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Ranking of Tennis Players http://sports.espn.go.com/sports/tennis/rankings?sport=WOMRANK

Page 30: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Uses of ordinal dataYou can express if something is greater than or less than (but you can’t express how much greater than or less than).

< > • Strongly agreeing is more than simply agreeing• Being tenth in the class is lower than being ninth

How much lower?

Page 31: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

IssuesThere are limitation of what you can do statistically

with ordinal data

Degree of violation Consensus

Page 32: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

NOIR Interval ScalingSatisfies all the requirements of an ordinal scale

(there is a high to low structure to the scale) and

• The intervals between the points on the scale become meaningful because the distance between successive points on an interval scale are equal.

Page 33: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Examples• Degrees Fahrenheit or Celsius• Calendar years

Interval scales do not have a meaningful zero point. They may contain a zero point, but it is arbitrary.

Page 34: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

NOIR Ratio• Satisfies all the requirements of an interval scale• There is a real and meaningful zero point on a

ratio scale• Weight, height, heart rate, breaths per minute,

degrees Kelvin, annual income, miles per hour, pulse, etc.

HINT: If your scale has negative numbers, like with temperature, then it is interval but it probably isn’t ratio.

Page 35: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Our Survey

R N O N N R N O O R

Page 36: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Measures of Central Tendency and Dispersion

Describing data

Page 37: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Describing Data vs. Describing People

Page 38: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Measures of Central Tendency

•Mean (M or X)•Median (Md)•Mode (Mo)

Page 39: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Issues with Describing Data1 2 4 61 2 4 61 2 4 6 n = 341 2 4 6 Mean = 3.15 ???1 2 42 3 42 3 4 What does that mean?2 3 42 3 52 4 5

Page 40: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

The MeanMost common synonym is the “average”

But what does the mean “mean”?

Definitions: • “arithmetic average”• a descriptor of the center of the data, when data

are distributed “normally”

Page 41: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

When a mean value is most useful• To simply summarize a data set that is normally

distributed • To summarize data from a sample that can be used

to estimate information about an entire population

Page 42: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

When a mean can be less useful as s descriptor

• To summarize data that is skewed (especially when it used as the only descriptor)

• To summarize data where there is an outlier

• To summarize data measured using a nominal scale

Page 43: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Mean and Skew- Find the mean annual income

$ 7,200$ 9,011 $ 20,074$ 24,999$ 36,567$ 32,145$ 54,158$567,987

$94,018How well does this number represent a measure of the center of our data set?

Page 44: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart
Page 45: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Mediana.k.a. Middle score in a ranked set of scores.It divides the distribution of scores into equal halves.

When there is an odd number of scores:1 2 2 3 5 5 7 8 10 15 16 16 21

When there is an even number of scores:4 5 7 15 16 19 31 32

Average = (15+16)/2 = 15.5 Median Score

Page 46: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Mean and Skew- Find the mean annual income

$28,572$7,200

$9,011

$20,074

$24,999

$32,145

$36,567

$54,158

$567,987

How well does this number represent a measure of the center of our data set?

10000.00 20000.00 30000.00 40000.00 50000.00

VAR00002

0.0

0.5

1.0

1.5

2.0C

ount

Page 47: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

ModeMost frequently occurring score is a group of scores.

Exam scores:

78 85 92 55 87 85 98 84 71

88 85 78 65 99 100 85 62 100

Page 48: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Measures of central tendencyGuide to which measures of central tendency are appropriate to use

with each scale of measurement:Mean Median Mode

Nominal XOrdinal (X)? X XInterval X X XRatio X X X

Page 49: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Measures of central tendencyGuide to which measures of central tendency are appropriate to use

with each scale of measurement:Mean Median Mode

Nominal XOrdinal (X)? X X

Interval X X XRatio X X X

Page 50: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Mean, Median and/or Mode

Page 51: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Class Survey Age VariableDescribe this dataset in terms of its

Mean

Median

Mode

Range

Descriptive Statistics

14 37 36.79 10.71414

ageValid N (listwise)

N Range Mean Std. Deviation

30 40 50

age

0

2

4

6

Cou

nt

Page 52: Deviations from NormalityNOIR R. atio • Satisfies all the requirements of an interval scale • There is a real and meaningful zero point on a ratio scale • Weight, height, heart

Class Survey Commute Variable

Describe this dataset in terms of its

Mean

Median

Mode

Range

What about this data set makes it difficult to describe using measures of central tendency?

10.0 20.0 30.0 40.0 50.0 60.0

travel

2

4

6

8

Cou

ntDescriptive Statistics

14 64.5 10.914 16.312414

travelValid N (listwise)

N Range Mean Std. Deviation

Commute

Giving more information, such as how skewed this variable is, would be helpful in the description.