Transcript Science and Research

Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering, and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
Introduction
• Data collected in quantitative research is in
the form of
– Numbers
• To use this data, researchers:
– Present it in charts or graphs
– Reorganize it for computer analysis
– Interpret or give theoretical meaning to it
Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering,
and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
Dealing with Data
• Coding - reorganizing raw data into a format that
– is easily entered into a computer
– or is machine-readable.
• Entering data – typically (see figure 12.1):
– each row is a case
– each column is a variable
– Four means of entering: code sheet, direct-entry, optical
scan, bar code
• Cleaning data
– checking the accuracy of coding and data entry.
Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering, and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
Descriptive Statistics
• Describe numerical data
– one variable at a time (univariate)
– two variables at a time (bivariate)
– or more than two (multivariate)
Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering, and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
Frequency Distributions
• Summarize information
– including counts and percentages
– and cumulative counts and percentages
– for nominal, ordinal, interval, or ratio
measurements.
• Graphic representations include the
– Histogram
– bar chart
– pie chart
Example of a histogram (showing two variables –
each bar would be a univariate histogram)
90
80
70
60
East
West
North
50
40
30
20
10
0
1st Qtr
2nd Qtr
3rd Qtr
4th Qtr
Example of a Pie Chart
1st Qtr
2nd Qtr
3rd Qtr
4th Qtr
Measures of Central Tendency
• Mode
– the most common or frequently occurring number.
• Median
– the middle point or 50th percentile
– used with ordinal, interval or ratio data
• Mean
– the arithmetic average used with interval or ratio level
data
– very sensitive to extreme values
Example of mean vs. median
We survey seven people and ask each how many alcoholic
drinks he or she consumed in the past month. The results
are
Person 1 2 3 4 5 6 7
Drinks 0 1 3 4 5 6 80
The median number is 4 – three people consumed fewer, and
three people consumed more
The mean number is 14.14: the total number of drinks is 99,
divided by 7 people is 14.4
From this example, you can see how ‘outliers’ – extreme
values – affect the mean much more than the median.
Measures of Variation
• Variation is
– the spread, dispersion, or variability
– around the center of the distribution
• Range
– the distance between smallest and largest scores
– e.g. ages might vary from a range of ages 21–59.
• Percentiles
– scores at a specific place within the distribution
– if someone age 26 is at the 25th percentile, that means
that 25% of the respondents were under age 26
Measures of Variation continued…
• Standard deviation
– an average distance of each score from the mean
– A nice explanation on the web
• Z score
– a standardized score
• What are standardized scores?
– it represents the number of standard deviations of a
particular score above or below the mean.
• One standard deviation away from the mean in either
direction on the horizontal axis (the red area on the above
graph) accounts for somewhere around 68 percent of the
people in this group. Two standard deviations away from
the mean (the red and green areas) account for roughly 95
percent of the people. And three standard deviations (the
red, green and blue areas) account for about 99 percent of
the people.
• If this curve were flatter and more spread out, the standard
deviation would have to be larger in order to account for
those 68 percent or so of the people. So that's why the
standard deviation can tell you how spread out the
examples in a set are from the mean.
Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering, and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
Results with Two Variables
• Bivariate statistics
– indicate whether there is a statistical
relationship between two variables
• There are two possible relationships:
– Covariation
• two variables are associated statistically.
– Independence
• there is no association between two variables
Seeing the relationship – the scattergram
• a graph on which a social researcher plots
each case or observation
• each axis represents the value of one
variable
What can be learned from a scattergram?
• Form - relationships can take three forms:
– independence (no relationship)
– linear (forming a straight line)
– curvilinear (forming either a ‘u’ or an ‘s’ curve).
• Direction - can be one of two values
– positive, higher values on one variable go with higher
values on the other
– negative, higher values on one variable go with lower
values on the other.
What can be learned continued…
• Precision
– the amount of spread in the points on a graph
– A high amount of precision occurs when the points hug
the line that summarizes a relationship,
– a low level of precision occurs when the points are
widely spread out.
Bivariate Contingency Table
• presents the same information as a
scattergram but in a more condensed
fashion.
• is ordinarily based on a cross tabulation of
two variables at the same time.
• Shows how the pattern of distribution of
one variable is “contingent” on the other
variable
Percentage Tables
• Be able to read a percentaged bivariate
contingency table, such as table 12.1 on
page 347
• Understand the difference between what is
shown in a
– Column-percentaged table
– Row-percentaged table
Reading a Percentage Table – Look At:
•
•
the title, variable names, and any background
information.
the direction in which percentages have been
computed, in rows or columns.
– How do you tell?
• See where the percentages total 100% (or near 100%)
•
the comparisons relevant to the cross tabulation.
–
–
Comparisons are made in the opposite direction
from that in which percentages are computed.
Compare across if the table is percentaged down,
compare down if percentaged across.
Example from the text
• Table 12.1, page 347
Measures of Association
• A measure of association is a single number that expresses
the strength, and often the direction, of a relationship
between two or more variables.
– It can help you interpret the pattern of data found in a bivariate
contingency table
• Researchers may choose from several different measures
of association
– The appropriate one depends partly on the level of measurement of
the variables (nominal, ordinal, interval, or ratio)
• Measures of association are lambda, gamma, tau,
chi (squared), and rho.
• If there is a strong association it means that there is a
definite pattern in predicting scores on the dependent
variable from variations in the independent variable.
Measures of Association continued…
• If there is a weak association it means that there is
not much of a pattern between scores on the
dependent variable compared to variations in the
independent variable.
• Measures of association normally range from 0.0
to +1.0, or from –1.0 to 0.0 to + 1.0.
• In either case, the closer the association is to 1.0
(+ or -), the stronger the relationship is
• The closer to 0.0, the weaker the association.
Measures of Association, continued
• Most measures of association follow a
“proportionate reduction in error” logic:
– How much does knowing the value of the
independent variable, for each case, help in
predicting the value of the dependent variable
– The better the prediction, the greater the
reduction in error
Five Measures
• Lambda is for nominal level data and
ranges from 0.0 to 1.0
• Gamma is for ordinal level data, and it
ranges from – 1.0 to 0.0 to +1.0
• Tau is for ordinal data, and is similar to
Gamma’s range of –1.0 to 0.0 to +1.0
Five Measures continued…
• Rho is Pearson’s Product Moment Correlation,
–
–
–
–
–
ranges from –1.0 to 0.0 to +1.0,
for data at the interval or ration level.
It is interpreted just like Gamma.
It can only measure linear relationships (not curvilinear)
It is the most commonly-used measure of correlation
• R-squared – the commonly-used term for Rho-squared:
– Tells what percentage of the variation in the dependent variable is
caused by the independent variable
• Chi Squared
– can be used as a measure of association in descriptive
statistics such as the others listed here
– or it can be used in inferential statistics to test a null
hypothesis.
– It ranges from 0.0 to infinity.
Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering, and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
Statistical Control
• A way to test whether an observed relationship
between two variables is spurious, which means:
– Caused by a third variable
– that separately affects the two variables we had been
examining
– Like in the examples we’ve seen:
• Ice cream consumption, short-sleeve shirts
– warm weather
• Use of night light, nearsightedness in children
– nearsightedness in parents
Statistical Control, continued
• New example from the text:
– Height and preference for baseball
• Taller children tend to like baseball more than shorter children
• What is the third variable here?
– Gender: affects both height (boys tend to be taller than girls) and
preference for baseball (boys tend to like baseball more than do girls)
• How does one “control” for a third variable?
– Essentially, by creating categories of the third variable, and testing
for the bivariate relationship within each category
– In this example, create two gender categories, male and female
– Ask whether:
• Taller boys prefer baseball more than do shorter boys
• Taller girls prefer baseball more than do shorter girls
– If the answers are no, then controlling for the third variable
eliminated the relationship between the first two variables
• This relationship turns out to be spurious
Statistical Control, continued
• When we look closely at such relationships,
by constructing trivariate tables, we may
find more complex results requiring more
complex explanations
The Elaboration Model of Percentaged Tables
• It is possible to create tables that include control
variables
• By creating separate subtables for each value of
the control variables
• In each subtable, we crosstabulate the independent
and dependent variables
• We will look at the case of one control variable
• Therefore we will be looking at trivariate tables
Example – based on text, tables page 352
• IV: concern for community
• DV: social action
• Control variable: sense of social justice
“Elaboration Paradigm”
• Each pattern represents a particular
combination of results, looking at:
– The bivariate table crosstabulating the IV and
DV, with no control variables
– The “partials” – each showing the
crosstabulation of the IV and DV for one value
of the control variable
Elaboration Paradigm, continued
• replication pattern
– partials show the same relationship between IV and DV
as does the bivariate table
– Therefore, the IV-DV relationship holds, even when
controlling for the third (control) variable
• specification pattern
– one partial replicates the initial bivariate relationship
but other partials do not.
– Therefore the IV-DV relationship holds, but only for
one value of the control variable
Elaboration Paradigm continued…
• Interpretation and explanation patterns
– The bivariate table shows a relationship between IV and DV
– But none of the partials tables show a relationship
– Conclusion: once the third variable is controlled for, the IV-DV
relationship disappears
– Therefore the apparent IV-DV relationship is
• Spurious
• The difference between interpretation and explanation is a
matter of whether the control variable comes before the IV
(explanation) or after the IV (interpretation)
Elaboration Paradigm continued…
• Suppressor pattern
– bivariate table: no relationship between IV and
DV
– Partials all (both) show IV-DV relationship
– But in opposite directions, so they cancel each
other out when combined into the bivariate
table
– Conclusion: IV and DV are associated; control
variable determines direction of relationship
A note on percentage tables
• Neuman’s tables – box 12.6, page 354 do not specify what
type of percentaging is being shown
– Cell: number of cases in cell / total number of cases
• four cells together total 100%
– Row: number of cases in cell divided by number in row
• The two cells in each row total 100%
– Column: number of cases in cell divided by number in column
• The two cells in each column total 100%
• A good explanation on the web
• Therefore, the bivariate table for the specification pattern
can exist, if the table is using ____ percentaging
– Column
• However, this would not match the partials shown
• Therefore, this is a mistake, and the table should look like
85 15
15 85
Multiple Regression Analysis
• A statistical technique for variables measured at
interval or ratio levels
• Results in a measure called R2 (R-squared), which
measures the combined influence of multiple
independent variables on one dependent variable
• Regression also shows the independent effect of
each variable, controlling for the other variables
• The effect on the dependent variable is measured
by a standardized regression coefficient: beta (ß)
• Example – see box 12.7, and associated text pp.
355-356
Chapter 12: Analysis of Quantitative Data
• Introduction
• Dealing with Data: Coding, Entering, and Cleaning
• Descriptive Statistics
– One Variable
– Two Variables
– More than Two Variables
• Inferential Statistics
• Conclusion
The Purpose of Inferential Statistics
1. Test hypotheses
(using probability theory)
2. Determine how confident one is in making
inferences from a sample to a population
3. Test whether descriptive results are likely to
be due to:
- real relationship, or
- random factors
Statistical Significance
• Means that results are unlikely to be due to
chance
• Indicates the probability of finding a
relationship in the sample when there is
none in the larger population.
• Cannot tell us if something is causing
something else
– it can only tell us what is likely.
Levels of Statistical Significance
• Statistical significance is usually expressed
in terms of levels
– usually .05, .01, or .001
• This means that results (within the sample)
are likely due to chance factors
– only 5%, 1%, or 1/10 % of the time,
• See different ways to express this, page 357
Type I and Type II Errors
• Type I Error
– is claiming that a relationship exists, when it does not
– In other words, falsely rejecting null hypothesis
• Type II Error
– Is claiming that there is no relationship in the data when
there really is one
– Is falsely accepting a null hypothesis
• Setting a very low acceptable significance level
(e.g. .001), increases the chances of type II error
• Setting a very high acceptable significance level
(e.g. .1), increases the chances of type I error
Type I and Type II Errors continued…
• The odds of making a Type I or a Type II
error are inversely proportional
• As the odds of making a Type I error
increase, the odds of making a Type II error
decrease.