STAT 366
Sample Survey Theory and Methods II
Lecture 1: Errors in Surveys
Dr. S.K. Appiah
Department of Mathematics
[email protected]/
[email protected]Nov 2016 020 527 9926
Course Contents:
1. Introduction:
• Review of random sampling
• Errors in surveys
2. Stratified sampling
3. Systematic sampling
5. Cluster sampling
6. Mini project
1
Assessment
1. Class test/Assignments (10 marks)
2. Mid-semester examination (20 marks)
3. End of Semester examination (70%)
2
Assignment 1: Mini Project (by Group Work)
• Select a topic of your choice to solve a real-life
problem which will require collection of data to be
analysed.
• Present your report following the format below:
1. Topic
2. Introduction:
• Brief background, problem statement and
objectives
3
3. Methodology:
• Selection of sample
• Data collection
• Analysis of data
4. Presentation of results and interpretations
5. Conclusion
• Summary of findings
• Justification and Recommendation(s)
6. References
4
1. Introduction
1.1 Basic concepts of random sampling:
Random selection of sample with known probability.
• Random sampling design serves as protection against
selection bias.
• Random sampling design makes the sample more
representative.
• Theoretical bridge-work of sampling theory rests on
randomness.
5
• Selection Methods of random sampling:
o Simple random sampling
(discussed last semester)
o Stratified sampling
o Systematic sampling
o Cluster sampling
o Multi-stage sampling
6
1.2 Errors in Surveys
• Two types of errors are observed in surveys:
o Sampling errors
o Non-sampling errors
• Sampling Errors:
o These errors are random which cannot be avoided.
o They arise solely from the measured n sample units
randomly selected from the N population units.
o Magnitude of the sampling error is determined by
standard error of a statistic.
7
• Sampling error measures the variability in all possible
samples
• Sampling error is used to calculate confidence
intervals.
• Sampling errors are reduced if sufficiently large sample
size (n) is obtained.
8
Non-Sampling Errors (NSEs)
• NSE is any phenomenon other than sampling error
which leads to biasedness.
• NSEs errors do occur at all stages of planning and
conduct of a survey.
• They pose danger to the survey results as their
magnitude is much larger
• There is no satisfactory technique of measuring the size
of these errors.
9
Causes of non-sampling errors
• Sampling operations
• Lack of knowledge of survey concepts
• Concealment of truth
• Problems of non-response
• Operationalization of Concepts/Conceptual Problems
• Improper design of questionnaire
• Measurement error
• Wrong data analysis and interpretation of results
• Lack of adequate supervision of fieldwork
10
• Sampling operations:
o Mistakes in drawing sampling units and/or
o Omission of part of the population from the sampling
frame.
• Lack of knowledge of survey concepts:
o Respondents may not understand information being
requested.
o They may also not know the information being
requested and do not try to ascertain the correct
information.
11
• Concealment of truth:
o Fear (probably of tax/revenue/levy)
o Suspicion or mistrust.
• Operationalization of Concepts:
o The translation of concepts used in a survey into
operation may be a problem.
• Problems of non-response:
o Non-coverage: failure to locate or visit some units in the
sample due to:
Poor weather condition/transportation facilities
Incomplete sampling frame or missing units;
12
o Inability to answer because the needed information to
certain questions may be unavailable or are unwillingly
given.
o Non-availability of up-to-date sampling frame.
o Refusal by “hard core” respondents (i.e. persons who
adamantly refuse to be interviewed for fear of tax,
being incapacitated).
o Absence from home during the survey period.
13
o Effects of non-response on survey results:
Reduction of number of units (n) to be covered and
amount of data for analysis, thus increasing the
standard error.
Prediction based on the estimates would be
insufficient and misleading.
Biasedness in the analysis as characteristics of non-
respondents may not have the same characteristics
as those who will respond.
14
• Improper design of questionnaire:
o Lengthy questionnaire for collection of needed data
may bore respondents.
o Wording of questions in the questionnaire may
influence respondents to answer in a specific way.
• Measurement error:
o Measurements made on units could be so biased
(e.g., an interviewer can rephrase the questions to
twist the responses in his/her own words, or record
wrongly)
o Devices for taking some measurements may be
faulty.
15
• Wrong data analysis and interpretation of results:
o Errors due to coding and keying in the data onto a
computer.
o Computer operations are subject to human errors.
o Wrong choice of methodology and/or wrong
interpretation.
• Lack of adequate supervision of fieldwork:
o May lead to serious mistakes in the data collection.
o Loss or damage of completed interview schedules
and other survey materials.
16
Control of Non-sampling errors.
• Proper design of surveys:
The best general method of controlling non-
sampling errors. This is achieved through:
o Careful and “error conscious” approach to the
design of survey procedures and documents.
o Good management of the survey (training of staff to
minimize interview errors).
17
• Reduction of non-response rate by substitution:
o Use of statistical methods to predict non-responses.
o Motivation and persuasion of both field staff and
respondents;
o Obtaining an up-to-date sampling frame;
o Pre-testing the questionnaire in a pilot-survey to
know more about the response units.
18
• Effective data collection and analysis:
o Good design of questionnaire
o Proper inspection of completed schedules.
o Precautions against mistakes and accidents in
handling survey instruments/materials.
o Appropriate method of data analysis
19
• Pilot-Survey:
o In a pilot survey, a small sample of the population is
enumerated for a major statistical investigation to be
undertaken.
o This provides valuable information about errors
which are likely to occur in the main survey.
o This tends to reduce errors in the main survey.
20
• Post-enumeration survey (PES):
o A PES is conducted immediately after the main
survey.
o It provides information on where errors may have
occurred.
o A PES normally consists of the re-enumeration of a
small sample of the population and the results
compared with those obtained originally.
o Usually, PES should be independent of the main
enumeration.
21
Thank You
For any concerns, please contact
[email protected]
[email protected]
0322 191132
Nov 2016
