Transcript No Slide Title

Hite study

Women in love: a cultural revolution in
progress, 1987, Shere Hite



84% of women not satisfied with their
relationships
70% of all women married >5 years have
extramarital affairs
95% of women report psychological and
physical harassment from their partners
1
Controversy


Widely criticized by media – “dubious,”
“of limited value”
Why?


Survey design (sampling methods,
questionnaire) inadequate
Did not lead to a survey data set that
supports inference to entire population of
women in US
2
Hite’s survey design

Sample

Addresses from broad range of special groups 
excludes many women in population  sampling
frame bias


Mailed questionnaires to 100K  4.5% returned
 low response rate (nonresponse bias)
Questionnaire



127 essay questions  high respondent burden,
nonresponse bias (who completes?)
Question wording vague (“in love” has many
different interpretations)  measurement error
Leading questions  response bias
3
Survey process
SURVEY DESIGN
COLLECT & PREPARE DATA

Define objectives & desired analyses

Collect data (interview,

Define target population
observe, self-administer)

Select sampling frame

Edit and code data

Choose sampling design, analysis approach  Enter data (if paper)

Choose data collection method

PREPARATION

Create sampling frame

Select sample

Develop questions or measurements

Construct questionnaire or other data
collection form

Pre-test questionnaire & revise

Train interviewers, data gatherers
Edit data file
DATA ANALYSIS

Exploratory data analysis

Calculate estimates of
population characteristics

Make inferences about the
population
4
Design for sample surveys


Survey design involves selecting methods for all
phases of the survey process, including sampling and
estimation
Sample design driven by




Objectives
Type of measurements to be taken (questions,
field observations)
Operational constraints ($, time, people, materials)
Analysis approach driven by



Objectives
Sample design (like design of experiments)
Data collected during the survey
5
Survey statistics

Study population
 Finite number of units
 1.7 million people in Nebraska
 18,567 students at UNL
 3000 counties in the US
 400 accounts being audited in a private firm
 Finite # of values  discrete distribution
6
Survey statistics - 2

Design


Very similar design structures
More explicit consideration of resource constraints
and analysis objectives than in experimental
design


Use stratification to obtain sufficient sample sizes for
subpopulations
Use cluster sampling to reduce costs of collecting data
7
Survey statistics - 3

Design-based estimation (this class)

Focus on estimating descriptive parameters:
means, proportions, totals



Less emphasis on regression, etc.
Based on randomization theory
Other approaches exist


Model-assisted (cover this a bit)
Model-based (not covered)
8
Definitions

Observation unit (OU)



Individual (student, animal, female), household, land area,
business, commercial account
May have more than one OU (cluster sampling later in
semester)
Target population




Students at UNL, US households, farms, forests
Impacts survey design and inferences that can be made
from survey
Can be hard to define
Political poll: are we interested in registered voters, voters
in last election, eligible voters?
9
Definitions - 2

Sample



Any method of selection (probability, quota,
volunteer)
We will focus on ways of selecting a sample that
use probability sampling
Sampling unit (SU)


May not be the same as the OU
Cluster sampling


OU = individual, SU = household
OU = elementary student, SU = school
10
Definition - 3

Sampling frame


Want this to at least include the entire target population
Some parts of frame may be outside the target population


Randomly selected telephone numbers include non-working
numbers that do not correspond to households
Sampled population – set of all possible OUs that
might have been chosen in a sample, or population
from which sample is selected


Ideally very close to target population
Does not include portions of target population that were


not sampled
sampled but failed to respond
11
Telephone survey of likely
voters (Fig 1.1, p. 4)

OU

Target pop

SU

Frame

Sampled population = ?
12
National Crime Victimization
Survey (NCVS)

Ongoing survey to study crime rates

Interested in total number of US households that
were victimized by crime last year

OU

Target population

Sampling frame

Sampled population
13
Pesticide survey

Survey of nitrate and pesticide contamination
in US drinking water
Target population

OU

Sampled population

14
What do we know about Hite’s
study?





OU
Target population
SU
Sampling frame
Sampled population
15
Selection bias


Occurs when some part of the target
population is not in the sampled population
May be due to ...



Sampling process
Data collection process
Can induce bias in estimated population
parameters

Bias occurs when the omitted part of target
population is different from the sampled
population with respect to the analysis variables
16
Types of selection bias
(Things you should avoid)

Convenience, volunteer samples

Take whomever is willing



Volunteer web surveys
Call-in surveys from TV programs
Judgment, purposive, quota samples


Select OUs without a probability mechanism
Pick sample using your judgment to reflect the target
population composition



Find a point on the land that “represents” a “typical” soil
condition
Mall intercept surveys may have a quota scheme
May be useful for initial studies to probe a topic

CANNOT make inferences about a population from such studies
17
Types of selection bias - 2
(Things you should avoid)

Ad hoc substitution of observation unit


If respondent not home, go to (unselected)
neighbor
Characteristics of substitute are likely to vary, may
alter sample composition
18
Types of selection bias - 3
(Things you can partially control)

Undercoverage – sampling frame omits portion of
target population



Homeless in telephone survey of U.S. residents
Unmapped waterways when sampling from USGS
topographic maps
Remedies

Select / construct sampling frame carefully



Cover as much of the target population as possible
Better if portion not covered by frame is small, or if it
differs in a way that minimizes impact on inferences
Once you have a frame, use probability sampling

Key to avoiding problems associated with convenience and
purposive samples
19
Types of selection bias - 4
(Things you can partially control)

Nonresponse during measurement process

Refusals



Not reachable


Can’t locate sampled person due to outdated contact info
Incompetent


Unit (refuse participation in survey)
Item (refuse to answer a question)
Too ill to complete survey, mentally/physically disabled
Remedies

Use multiple and persistent methods to find / reach OU



Variety of address sources (web, change-of-address)
Multiple attempts to call at different times of week / day
Use rigorous methods encourage OU to participate

Refusal conversion techniques, incentives, rapport (see later)
20
1936 Literary Digest survey

Predicted correctly presidential election outcome 19121932


Used “commercial sampling methods” used to market
books



1932: Predicted Roosevelt w/ 56%, got 58% in election
Telephone books, club rosters, city directories, registered voter
lists, mail-order lists, auto registrations
Mailed out 10 million questionnaires, received 2.3 million
1936


Predicted Roosevelt loss (41% to Landon’s 55%)
Roosevelt won, 61% to 37%
21
What happened?

Undercoverage in sampling frame



Low response rate



Heavy reliance on auto and phone lists
Those w/ cars and/or phones voted in favor or
Roosevelt, but not to the extent that those without
cars and phones did
Those responding preferred Landon relative to
those who didin’t
Many Roosevelt supporters didn’t remember
receiving survey
Large sample is no guarantee of accuracy
22
Selection bias
nearly always exists

Want sample and resulting survey data to be
“representative” of the target population


Methods should be described in documentation and
published articles


Good survey design and proper implementation of protocols
are key to minimizing selection bias
Enable user/reader to make judgments about the nature of
selection bias and its effects on the interpretation of results
Useful to explicitly define the sampled population to
reflect selection bias that has occurred in the survey
process

Likely voters with telephones who could be reached and
were willing and able to respond to the survey
23
Measurement bias


Ideally, want accurate responses to questions
or measurements of phenomena
Measurement bias occurs when measurement
process produces observations on an OU that
differ from the true value for the OU in a
systematic manner



Calibration error in scale adds 5 kg to weight for
each person in a health survey
Bird surveys record species heard or sighted in 0.5
km radius during a 10 min period
Fail to present a valid option in a response list
24
Measurement bias in people

Respondent may provide false information




More likely with sensitive subject matter
Socially acceptable behavior (drug use)
Desire to influence outcome of survey to reap
benefit (ag yields)
Memory


Recall bias – distant memory more prone to error
Telescoping – recall events that occurred before
reference period
25
Measurement bias in people - 2

Impact of interviewer

Respondent reactions


Caucasians provide different answers to white
and black interviewers, vice versa
Interviewer interaction with respondent


Misreading questions
Poor rapport
26
Measurement bias in people - 3

Impact of questionnaire

Respondent fails to understand question


May not understand terms, be confused by question, not
hear correctly
Variation in interpretation of of words or phrases


Even simple questions may not be explicitly clear
Do you own a car?



Is “you” singular or plural?
Is a van or truck included in the concept of a car?
Question order


Context effects – previous question impacts answer
Poorly organized questionnaire can make it difficult for
respondent to understand questions
27
Questionnaire design

Clearly and specifically define study
objectives




Specific topics and questions for study
Identify target (sub)populations and contextual
variables for analysis (e.g., demographics)
Evaluate proposed questions as to whether
they clearly support objectives and analysis
methods
Pre-test the survey instrument (=questionnaire)


On respondents from the target population
Large-scale surveys may rely on intensive study

NCVS: alternative recall periods, question wording
28
Writing questions


Use clear, simple, precise language
Focus on one well-defined item in a question





Avoid referring to multiple concepts in a single
question
Divide lengthy questions into a contextual
statement plus a simple question
Specify a time frame, area, or other form of scope
Define critical terms
State question neutrally

Avoid leading questions that might induce bias
29
Writing questions - 2

Response formats



Use mutually-exclusive categories in closed-ended
questions
Reduce post-hoc coding by minimizing use of
open-ended questions
Organization


Group questions to improve ability of respondent
to follow content and understand questions
Put key questions first while the respondent is
fresh (but start easy)
30
Impact of measurement bias

Measurement bias via data collection
procedures


Individual observation level
Bias at the observation level impacts
estimates in two ways


Systematic bias over OUs in sample in same
direction results in a biased estimate of a
population characteristic
Measurement error often results in increased
variance in estimates (with or without bias) as
well
31
Nonsampling Errors
(Lessler & Kalsbeek, 1992)

Assume: probability sample

Frame error



Mismatch between sampled population & target
population
Nonresponse error

Unable to obtain data from observation units

Whole observation unit or single response item
Measurement error

Inadequacies in the process of obtaining
measurements from observation units
32
Survey error model
Total
Survey
Error
Assessed via
bias and
variance
=
+
Due to the
sampling
process (i.e.,
we observe
only part of
population)
Measurement error
Nonresponse error
Frame error
33
Sampling Error

Sample survey



Collecting data from a sample – a subset of the population –
to make inference about the whole population
We never observe the whole population  estimate for any
one sample is unlikely to perfectly match the population
parameter
Example




Proportion of undergraduates in Fall 2000 that are males =
44.6%
Select a sample of 100 undergrads  estimate = 46.2%
Select a sample of 100 undergrads  estimate is 41.9%
Etc.
34
Why sample?

Widely accepted that sample surveys of large
populations will lead to more precise
estimates than a census of the population


Sampling error vanishes, but measurement error is
typically much higher
US example



Number of occupied housing units (N) = 105,480,101
Federal statistical survey sample size (n) = 50,000
May not be a need to select a sample with
small populations (e.g., web or mail surveys)


Membership of organizations
Employees in a business
35