A Statistical Sampler
To understand God's thoughts we must study statistics, for these are the measure of His purpose. — Florence Nightingale
Statistical Terms Crossword
To behold is to look beyond the fact; to observe, to go beyond the observation. Look at the world of people, and you will be overwhelmed by what you see. But select from that mass of humanity a well-chosen few, and observe them with insight, and they will tell you more than all the multitudes together. — Paul D. Leedy From his book, “Practical Research,” 1993
Choosing the Appropriate Statistic Some factors to consider:
• Research design • Number of groups • Number of variables
• Level of measurement (nominal, ordinal, interval/ratio)
Statistical Methods
Statistical Methods Descriptive Methods
Inferential Methods Univariate
Applied to means
Bivariate
Applied to other statistics
Multivariate
Descriptive Statistics
Descriptive Methods Univariate
Bivariate
Multivariate
shape
correlation
spread
regression
multiple regression
Inferential Statistics
Inferential Methods Applied to means
Applied to other statistics 2 groups: t-test
>2 groups: ANOVA
While the individual man is an insoluble puzzle, in the aggregate he becomes a mathematical certainty. You can, for example, never foretell what any one man will be up to, but you can say with precision what an average number will be up to. Individuals vary, but percentages remain constant. So says the statistician. — Arthur Conan Doyle
Some Statistics-Related Web Sites The University of Kansas Virtual Statistical Assistant http://www.ku.edu/~coms/virtual_assistant/vsa/
Biostatistics for the Clinician Hypertext Glossary Part 1: http://www.uth.tmc.edu/uth_orgs/educ_dev/oser/LGLOS1_0.HTM Part 2: http://www.uth.tmc.edu/uth_orgs/educ_dev/oser/LGLOS2_0.HTM
Research Methods Knowledge Base http://www.socialresearchmethods.net/kb/
Types of Statistics •
Descriptive statistics characterize the attributes of a set of measurements. Used to summarize data, to explore patterns of variation, and describe changes over time.
•
Inferential statistics are designed to allow inference from a statistic measured on sample of cases to a population parameter. Used to test hypotheses about the population as a whole.
Requisite Conditions for Causation In order for X to cause Y: • X & Y must be associated
• X must precede Y in time • X contains unique information about Y that is not articulated elsewhere
The invalid assumption that correlation implies cause is probably among the two or three most serious and common errors of human reasoning.
— Stephen Jay Gould, The Mismeasure of Man
Smoking is one of the leading causes of statistics. — Fletcher Knebel
Randomization • Random selection is how you draw the sample for your study from a population. • This is related to the external validity, or generalizability, of your results.
Randomization • Random assignment is how you assign your sample to groups or treatments in your study. • This is related to internal validity. • Random assignment is a required feature of a true experimental design.
Randomization
Variables • Variables are qualities, properties, or characteristics of persons, things, or situations that change or vary and are manipulated, measured, or controlled in research. • More simply stated: Variables are things that we measure, control, or manipulate in research.
Types of Variables • Independent variables are manipulated or varied by the researcher, for example, intervention or treatment. • Dependent variables are the responses, outcomes, etc. that are measured by the researcher. • Extraneous variables are not part of the research design, but may have an impact on the dependent variable(s).
Levels of Measurement • Nominal • Ordinal • Interval • Ratio
Nominal-Level Variables • Data are organized into categories • Categories have no inherent order
• Categories are exclusive • Categories are exhaustive • Examples are sex, ethnicity, marital status
Examples of Nominal-Level Questions • Do you have a loss of appetite?
• Do you smoke a lot? • What is your ethnicity?
Ordinal-Level Variables • Categories can be ranked in order • Intervals between categories may not be equal • Examples are socioeconomic status, level of education attained (elementary school, high school, college degree, graduate degree)
Examples of Ordinal-Level Questions •
Would Intervention X be your 1st, 2nd, or 3rd choice of treatment for Condition Y? 1 First choice 2 Second choice 3 Third choice
•
Beck Depression Scale – Sadness Item 0 I do not feel sad 1 I feel sad 2 I am sad all the time and I can’t snap out of it 3 I am so sad or unhappy that I can’t stand it
Interval-Level Variables • Distances between levels of the scale are equal • Assumed to be a continuum of values • An example is temperature (measured in Fahrenheit or Centigrade)
Examples of Interval-Level Variables • IQ scores • GRE scores • Composite scores of multi-item scales
Ratio-Level Variables • Equal spacing between intervals • Have an identifiable absolute zero point • Examples are weight, length, volume, and temperature (measured in Kelvin) • In statistical analysis, typically there is no distinction made between interval level and ratio level
Same Variable, Different Levels of Measurement Interval level: What is your age in years?
Ordinal level: What is your age group? 18 years or younger 19-44 years 45 years or older
____
Importance of Levels of Measurement • Level of measurement is associated with the type of statistical method used. • Higher levels of measurement provide more information than do lower levels. • In general, you should use the highest level of measurement possible. For example, measure actual age in years, not in age groups.
Some Major Types of Analyses • Description • Relationships among variables • Differences between groups or treatments
There are three kinds of lies – lies, damned lies and statistics. — Benjamin Disraeli
Measures of Central Tendency Level of Measurement
Statistic
Nominal
Mode
What is the most frequent value?
Ordinal
Median
What is the middle score? (50% above and 50% below)
Mean
What is the average? (Sum of all scores divided by the number of scores)
Interval/Ratio
Example of Central Tendency
15,20,21,20,36,15,25,15 15,15,15,20,20,21,25,36
Example of Mode Race of Respondent
RACE Race of Respondent 1400
1 white 2 black 3 other Total
Frequency 1257 168 75 1500
Percent 83.8 11.2 5.0 100.0
Statistics
1000 800 600 400
Frequency
RACE Race of Res pondent N Valid 1500 Mis sing 0 Mode 1
1200
200 0 w hite
Race of Respondent
black
other
Example of Median EDUC Education level
4 Some high s chool 5 Completed high school 6 Some college 7 Completed college 8 Some graduate work 9 A graduate degree Total
Frequency 1 6 6 3 4 4 24
Percent 4.2 25.0 25.0 12.5 16.7 16.7 100.0
10
Cumulative Percent 4.2 29.2 54.2 66.7 83.3 100.0
9
8
7
6
Statistics EDUC Education level N Valid Mis sing Median
5
24 0 6.00
4 3 N=
24
Education level
Example of Mean Age of Respondent 200
MEAN
100
Std. Dev = 17.42 Mean = 46 N = 1495.00
0 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
Age of Respondent
I abhor averages. I like the individual case. A man may have six meals one day and none the next, making an average of three meals per day, but that is not a good way to live. — Louis D. Brandeis
Measures of Variation Level of Measurement
Statistic
Nominal
Number of categories
How many different values are there?
Ordinal
Range
What are the highest and lowest values?
Interval/Ratio
Standard Deviation
What is the average deviation from the mean?
Curves of Distribution
Normal Distribution
Normal Curve
Example: Number of categories
Race of Respondent
RACE Race of Respondent 1400
Frequency 1257 168 75 1500
Percent 83.8 11.2 5.0 100.0
1200 1000 800 600 400
Frequency
1 white 2 black 3 other Total
200 0 w hite
Race of Respondent
black
other
Example of Range EDUC Education level
Frequency 4 Some high s chool 1 5 Completed high school 6 6 Some college 6 7 Completed college 3 8 Some graduate work 4 9 A graduate degree 4 Total 24
Percent 4.2 25.0 25.0 12.5 16.7 16.7 100.0
Cumulative Percent 4.2 29.2 54.2 66.7 83.3 100.0
10
9
8
7
Statistics 6
EDUC Education level N Valid Mis sing Median Range Minimum Maximum
24 0 6.00 5 4 9
5
4 3 N=
24
Education level
Example of Standard Deviation Age of Respondent 200
-1 SD
MEAN
+1 SD
Frequency
100
Std. Dev = 17.42 Mean = 46 N = 1495.00
0 20
30 25
40 35
50 45
Age of Respondent
60 55
70 65
80 75
90 85
Measures of Relationships
Level of Measurement
Statistic
Nominal
Phi statistic ()
Ordinal
Spearman rho () correlation
Interval/Ratio
Pearson correlation (r)
Statistics have shown that mortality increases perceptibly in the military during wartime. — Robert Boynton
Example of Spearman Correlation DEGREE RS Highest Degree
RINCOM91 Respondent's Income
Valid
Mis sing Total
1 LT $1000 2 $1000-2999 3 $3000-3999 4 $4000-4999 5 $5000-5999 . . . 19 $50000-59999 20 $60000-74999 21 $75000+ Total
Frequency 26 36 30 24 23 . . . 38 23 44 947 553 1500
Percent 1.7 2.4 2.0 1.6 1.5 . . . 2.5 1.5 2.9 63.1 36.9 100.0
Valid Percent 2.7 3.8 3.2 2.5 2.4 . . . 4.0 2.4 4.6 100.0
Valid
0 Les s than HS 1 High school 2 Junior college 3 Bachelor 4 Graduate Total
Mis sing Total Total
Frequency 279 780 90 234 113 1496 4 4 1500
Percent 18.6 52.0 6.0 15.6 7.5 99.7 .3 .3 100.0
Correlations
Spearman's rho
EDUC Highes t Year of School Completed
RINCOM91 Res pondent's Income Correlation Coefficient .363** Sig. (2-tailed) .000 N 945
**. Correlation is significant at the .01 level (2-tailed).
Valid Percent 18.6 52.1 6.0 15.6 7.6 100.0
Scatterplot of Self Esteem By Height
Relationship Between Two Variables Positive Correlation
Negative Correlation
Curvilinear Relationship
Example of Pearson Correlation • •
Variable HEIGHT is measured in inches Variable ESTEEM is the average of 5 items measured on a four-point scale (1-4) 4.0
Statistics
N
Valid Mis sing
Mean Std. Deviation
HEIGHT 24 0 66.7917 7.03395
ESTEEM 24 0 2.7583 .59558
3.5
3.0
Correlations
Pears on Correlation Sig. (2-tailed) N
2.5
ESTEEM
HEIGHT
ESTEEM .347 .097 24
2.0
1.5 50
HEIGHT
60
70
80
90
Example of Chi-Square Test RACE * SEX Crosstabulation SEX
RACE
Total
1 white 2 black 3 other
1 Male Count % within SEX 552 86.1% 66 10.3% 23 3.6% 641 100.0%
2 Female Count % within SEX 705 82.1% 102 11.9% 52 6.1% 859 100.0%
Total Count % within SEX 1257 83.8% 168 11.2% 75 5.0% 1500 100.0%
Chi-Square Tests
Pears on Chi-Square N of Valid Cas es
Value 5.994 a 1500
df 2
Asymp. Sig. (2-s ided) .050
a. 0 cells (.0%) have expected count less than 5. The minimum expected count is 32.05.
A Statistical Sampler
Take a 15 minute break!
Statistical thinking will one day be as necessary a qualification for efficient citizenship as the ability to read and write. — H.G. Wells
Some Terminology • Descriptive statistics Statistics that allow the researcher to organize or summarize data to give meaning or facilitate insight.
• Inferential statistics Methods that allow inferences to be made from a sample to a population
• Hypothesis testing A statistical test of an expected relationship between two or more variables
Statistical inference Statistical inference is the process of estimating population parameters from sample statistics.
Statistical inference may be used to ascertain whether differences exist between groups...
90
Height in inches
80 70 60 50 40 30 20 10
Males
Females
Are males taller than females?
... or whether there is a relationship among variables. SELF ESTEEM SCORE
4.0 3.5 3.0 2.5
GENDER 2.0 FEMALES 1.5 20
MALES 30
40
50
60
AGE
Is there a relationship between age and self-esteem? Does this relationship differ for males and females?
Examples of Some Commonly Used Statistical Tests Level of Measurement Nominal
Number of groups 1 group 2 independent groups
2
test
2 test
Ordinal
Interval/Ratio
Kolmogorov-Smirnoff 1 sample test
t-test of sample mean vs. known population value
Mann-Whitney U test
Independent samples t-test
2 dependent groups
McNemar test
Wilcoxon test
Paired t-test
>2 independent groups
2 test
Kruskal-Wallis ANOVA
ANOVA
>2 dependent groups
Cochran Q test
Friedman ANOVA by ranks
Repeated measures ANOVA
Some Commonly-Used Multivariate Methods
• Analysis of Variance and Covariance Tests for differences in group means • Multiple Regression Analysis Estimates the value of a dependent variable based on the value of several independent variables
Some Commonly-Used Multivariate Methods
• Reliability analysis Assesses the consistency of multi-item scales • Factor Analysis Examines the relationships among variables and reveals related sets of variables (constructs) • Structural Equation Modeling Methods for testing theories about the relationships among variables
Hypothesis Testing Decision Chart Reality
Null Hypothesis (H0 ) is true
Alternative Hypothesis (H1) is true
Type I error ()
Correct decision
typically .05 or .01
typically .80
Correct decision (1 - )
Type II error ()
typically .95 or .99
typically .20
Decision
Reject (H0 )
Don’t reject (H0 )
(Power = 1 - )
Difference between two group means: The independent samples t-test Males and females are asked a question that is measured on a five-point Likert scale: To what extent do you feel that regular exercise contributes to your overall health? 1 2 3 4 5
Strongly agree Agree Neither agree nor disagree Disagree Strongly disagree
Do males and females differ in their response to this question?
25 males and 25 females answered our question. Here is how they responded:
males females 1
2
3
meanmales=2.5 meanfemales=3.2
4
5
We can use the SPSS statistical package to run an independent samples t-test: First we enter the data into SPSS.
Then we invoke the Independent Samples T-Test procedure.
We tell SPSS which is the dependent variable and which is the independent variable to use in performing the t-test:
SPSS gives us summary statistics for each group: Group Statistics
EXERCISE
GENDER 1 male 2 female
N 25 25
Mean 2.56 3.24
Std. Deviation 1.158 1.012
Std. Error Mean .232 .202
The t-test reveals a significant difference between males & females: Independent Samples Test t-tes t for Equality of Means
EXERCISE
t -2.212
df 48
Sig. (2-tailed) .032
Mean Difference -.68
Reporting Results • See the guidelines in the APA Publication Manual, Fifth Edition • The manual provides very specific instructions for presenting statistical results. • Example:
The mean exercise score for females, 3.24, was significantly higher than for males, 2.56, t(48) = 2.12, p = .032.
Do the educational levels of males and females differ? 10
9 8 7 6 5 4 3 2 1
Education level
9
A graduate degree Some graduate work Completed college Some college Completed high school Some high school Completed grade school Some grade school No formal education
8 7 6 5 4 3 N=
14 Female
10 Male
Gender
Because the dependent variable (education level) is ordinal-level, we use the Mann-Whitney U Test. Ranks
For each group, the Sum and mean of ranks Is computed.
EDUC Education level
GENDER 1 Female 2 Male Total
N 14 10 24
Mean Rank 13.46 11.15
Test Statisticsb
The test statistics suggest that males’ and females’ education levels do not differ in this population.
Mann-Whitney U Wilcoxon W Z Asymp. Sig. (2-tailed) Exact Sig. [2*(1-tailed Sig.)]
EDUC Education level 56.500 111.500 -.807 .420 .437
a. Not corrected for ties. b. Grouping Variable: GENDER
a
Sum of Ranks 188.50 111.50
Difference between two groups over time: Repeated measures analysis of variance • Asthmatic elementary school children are given training intended to reduce the number of asthmatic episodes. • A control group is not given the training. • Children’s school attendance is monitored during the month before training is given to the intervention group, and during each of the two months following the intervention. • Does the asthma training intervention improve the school attendance relative to the control group?
The experimental design: Month 0
Intervention
Month 1
Month 2
Intervention Group
O
X
O
O
Control Group
O
O
O
O = observation
X = treatment/intervention
We can use the SPSS statistical package to perform a repeated measures ANOVA on the sample data:
First we enter the data into SPSS.
Then we request the General Linear Models procedure for Repeated Measures.
Here are the results involving time: Tests of Within-Subjects Effects Meas ure: ATTEND
Source TIME TIME * GROUP Error(TIME)
Type III Sum of Squares .034 .080 .244
df 2 2 24
Mean Square .017 .040 .010
F 1.695 3.956
Sig. .205 .033
The time x group interaction is significant.
And here are the results involving group: Tests of Between-Subjects Effects Meas ure: ATTEND Trans formed Variable: Average Source Intercept GROUP Error
Type III Sum of Squares 28.271 .068 .054
df 1 1 12
Mean Square 28.271 .068 .004
F 6293.102 15.201
Sig. .000 .002
The main effect involving group is significant.
This is a plot of the group means over time Estimated Marginal Means of ATTEND
Attendance (% of days)
100%
90% Intervention
Control 80%
70% Month 0
Month 1
TIME
Month 2
Factor Analysis Example The General Social Survey (GSS) is an “almost annual” personal interview survey of U.S. households conducted by the National Opinion Research Center. In the 1993 GSS, approximately 1500 adult respondents (18 years or older) were asked about their music preferences. Just for the fun of it, I performed a factor analysis on the music questions to see if we could identify a pattern of underlying dimensions, or factors, in the data.
MUSIC GENRES
I'm going to read you a list of some types of music. Can you tell me which of the statements on this card comes closest to your feeling about each type of music. (HAND CARD “B” TO RESPONDENT.)
Big Band Bluegrass Country/Western Blues or R & B Broadway Musicals Classical
Folk Jazz Opera Rap Heavy Metal
RESPONSE CARD “B”
Let's start with big band music. Do you like it very much, like it, have mixed feelings, dislike it, dislike it very much, or is this a type of music that you don't know much about?
1 2 3 4 5 8 9
Like Very Much Like It Mixed Feelings Dislike It Dislike Very Much DK Much About It NA
Factor Analysis Results The factor analysis revealed four factors in the music preference items. The varieties of music were associated with the factors as shown below: Pattern Matrixa Factor CLASSICL Clas sical Music OPERA Opera MUSICALS Broadway Mus icals FOLK Folk Music BIGBAND Bigband Music JAZZ Jazz Mus ic BLUES Blues or R & B Mus ic BLUGRASS Bluegrass Music COUNTRY Country Wes tern Mus ic HVYMETAL Heavy Metal Mus ic RAP Rap Mus ic
1 .844 .715 .663 .502 .459 .035 -.024 .070 -.084 -.012 .030
2 -.033 -.004 .109 -.064 .240 .766 .714 .084 -.034 -.016 .074
Extraction Method: Principal Axis Factoring. Rotation Method: Oblimin with Kaiser Normalization. a. Rotation converged in 8 iterations .
3 -.127 -.032 -.024 .341 .125 -.110 .106 .753 .596 .020 -.004
4 .054 .086 -.104 -.005 -.171 .029 .057 .052 -.033 .602 .559
Factor Analysis Results FACTORS
F1
Classical
F2
F3
F4
Folk
Musicals
Big Band
Opera
Jazz
Blues
Bluegrass
Country
MEASURED VARIABLES
Heavy Metal
Rap
Do not put faith in what statistics say until you have carefully considered what they do not say. — William W. Watt
More Cool Statistics Web Sites Rice Virtual Lab in Statistics http://www.ruf.rice.edu/~lane/rvls.html
Multimedia Resources for Statistics Students http://research.ed.asu.edu/msms/multimedia/multimedia.cfm
Statistics and Statistical Graphics Resources http://www.math.yorku.ca/SCS/StatResource.html
Without data, all you are is just another person with an opinion. — Unknown
Statistical Power Analysis • Prior to conducting a study, it is advisable to conduct a statistical power analysis.
• Power is the probability that a statistical test will detect a significant effect that exists. • The power analysis will suggest an adequate sample size for the study.
Four parameters related to the power of a test: • Significance level ()
• Sample size (n) • Effect size (ES) • Power (1 - )
Relationship between power and other parameters: • As significance level () decreases numerically, power decreases • As effect size increases, power increases • As sample size increases, power increases
Conventions commonly used:
Significance level ():
.05 * .01
.001 Effect size:
“small” “medium” * “large”
Power:
.80 * .90
*
Typical values for social/behavioral/health sciences
Examples of Effect Size:
EFFECT SIZE TYPE OF TEST
Independent Samples Ttest
Product Moment Correlation
MEASURE OF EFFECT SIZE SMALL
MEDIUM
LARGE
|mA-mB|
.2
.5
.8
rXY
.10
.30
.50
Testing a mean against a true alternative: 1 slightly larger than 0 (“small effect”) Sampling distribution of means when H0 is true
Area=
Sampling distribution of means when H1 is true
Area=1-
Area=
0 Region of nonrejection
1 Critical value
Region of rejection
Testing a mean against a true alternative: 1 quite a bit larger than 0 (“large effect”)
Area=
Area=1-
Area=
0 Region of nonrejection
1 Critical value
Region of rejection
Relationship Between Alpha(), Sample Size (n), and Power (1-) Two group t-test of equal means (equal n's) Æ α = 0.025 ( 2) Êδ = 0.500 α = 0.050 ( 2) Êδ = 0.500 Æ α = 0.100 ( 2) Ê Æ δ = 0.500
100
Power
90 80
power=.80
70 60 n=51
50 20
40
n=64
n=78
60 80 Sample Size per Group
100
120
The Power Analysis “Bible”
There are a lot of statistical power analysis resources (including interactive “power calculators”) on the World Wide Web. For example, see the StatPages.net web site at: http://members.aol.com/johnp71/javastat.html#Power Or, using a WWW search engine like Yahoo or Google, use the search string: statistical power analysis
Getting Help • For course assignments involving statistics, see your instructor or teaching assistant. • For help related to a masters thesis or applied project, see your faculty advisor. • Your instructor or advisor may confer with or make an appointment as needed with a statistician in the College of Nursing Center for Research and Scholarship.
Getting Help The Statistics Hotline is sponsored by a joint effort of the ASU Committee on Statistics, the Department of Mathematics and Statistics, and the Division of Graduate Studies. Its services are available to anyone affiliated with ASU and needs assistance with their ASU-related research.
http://www.asu.edu/graduate/statistics/hotline/
An approximate answer to the right question is worth a great deal more than a precise answer to the wrong question. — The first golden rule of mathematics, sometimes attributed to John Tukey
Statistical Terms Crossword Solution
On the Web
This presentation is available online in Microsoft PowerPoint format at: http://www.public.asu.edu/~eagle/stat_sampler.ppt