Data
facts or information that is relevant or appropriate to a decision maker
Population
the totality of objects under consideration
Sample
a portion of the population that is selected for analysis
Parameter
a summary measure (e.g., mean) that is computed to describe a characteristic of the population
Statistic
a summary measure (e.g., mean) that is computed to describe a characteristic of the sample
Data
facts or information that is relevant or appropriate to a decision maker
Population
the totality of objects under consideration
Sample
a portion of the population that is selected for analysis
Parameter
a summary measure (e.g., mean) that is computed to describe a characteristic of the population
Statistic
a summary measure (e.g., mean) that is computed to describe a characteristic of the sample
1. Gender = categorical - nominal
2. Weight = numerical - continuous - ratio
3. Auto Speed = numerical - continuous - ratio
4. Temperature = numerical - continuous - interval
5. # Siblings = categorical - ordinal
6. Letter Grade = categorical - ordinal
Question Content
Related to research purpose
Based on respondent’s ability to answer accurately
Response Format
open-ended Vs. fixed alternative (closed-ended)
Question Wording
simple, clear words
avoid leading questions: ‘In view of the health crisis, it would be best to nationalize the health industry.’
Question Sequence
use simple & interesting opening questions
general questions first
Layout
More important for Mail Survey than telephone survey
Pretest
Shows where you have asked ambiguous questions
Pragmatic Reasons
If Chrysler wished to census past purchasers’ reactions, millions of car buyers would have to be contacted
Accurate & Reliable Results
Reasonable accuracy though not perfect - sampling error!
May be more accurate than census since less chance of nonsampling errors (e.g., data entry)
Bureau of the Census uses samples to check the accuracy of the US Census. If the sample shows possible source of error, the census is redone.
Destruction of Test Units
e.g., Mean Life of Light Bulbs
Probability Samples
Selection is based on chance
Subjects are chosen based on some known probabilities
Eliminates or reduces bias
Random refers to procedure not the data:
The outcome cannot be predicted because it is dependent upon chance
Non Probability Samples
Do not have above characteristics
Done for time and convenience
Probability Samples
Selection is based on chance
Subjects are chosen based on some known probabilities
Eliminates or reduces bias
Random refers to procedure not the data:
The outcome cannot be predicted because it is dependent upon chance
Non Probability Samples
Do not have above characteristics
Done for time and convenience
Simple Random
Use random number table
Number of digits is determined by population size
Columns are 01, 02 etc. (aligned vertically)
Example
Population size is 50. Sample size is 10.
Since population size (50) has 2-digits, divide table into 2 digit numbers.
Begin top left (for convenience only).
1-49, 2-28, 3-08, 4-89 (skip) 4-24, 5-35, 6-77 (skip),
6-90 (skip) 6-02, 7-83 (skip) 7-61(skip) 7-87 (skip)
7-04, 8-16, 9-57 (skip) 9-07, 10-46.
Example
Population size is 100. Use 3 digit numbers.
Probability Samples
Selection is based on chance
Subjects are chosen based on some known probabilities
Eliminates or reduces bias
Random refers to procedure not the data:
The outcome cannot be predicted because it is dependent upon chance
Non Probability Samples
Do not have above characteristics
Done for time and convenience
Systematic
Requires all population elements
Bias may occur due to periodicity
In the telephone book example, unlisted numbers will not be found
Example:
Sampling frame is 100 individuals. You want to select 20. Select first name by random number, then every 5th person.
Probability Samples
Selection is based on chance
Subjects are chosen based on some known probabilities
Eliminates or reduces bias
Random refers to procedure not the data:
The outcome cannot be predicted because it is dependent upon chance
Non Probability Samples
Do not have above characteristics
Done for time and convenience
Stratified
Assures
1. Sample reflects population in terms of criterion used for stratifying.
2. More efficient sample - sampling error is reduced.
Example: College has 70% on-campus students and 30% commuters. A 100 student survey would get close to 70 on-campus students and 30 commuters. A simple random survey might get 60 on-campus and 40 commuting students.
Similar to Quota sampling except that a simple random sample is drawn from each strata.
Probability Samples
Selection is based on chance
Subjects are chosen based on some known probabilities
Eliminates or reduces bias
Random refers to procedure not the data:
The outcome cannot be predicted because it is dependent upon chance
Non Probability Samples
Do not have above characteristics
Done for time and convenience
Cluster
Idea is to sample economically yet retain characteristics of probability sample.
Ideally, cluster is as heterogeneous as the population.
Often, characteristics of elements in cluster may be similar.
Probability Samples
Selection is based on chance
Subjects are chosen based on some known probabilities
Eliminates or reduces bias
Random refers to procedure not the data:
The outcome cannot be predicted because it is dependent upon chance
Non Probability Samples
Do not have above characteristics
Done for time and convenience
Judgment
A fashion manufacturer selects key accounts to predict what will sell next season
Quota
Advantages are speed of data collection, lower costs, and convenience.
Often used in laboratory experiments
It is difficult to find a sample of the general population willing to visit a laboratory
Chunk (Convenience)
Street interviews at election time. Views represent supposedly the entire population.
Need impressions of text book in an hour. Use this class to represent all students.
Frame Error
The sampling frame is also called the ‘working population.’
Frame error is the discrepancy between population and sampling frame.
e.g., Not all students may be in phone book
Sampling Error
Sampling units may not perfectly represent the population.
All samples vary.
Sampling error is a function of sample size
Systematic (Nonresponse & Measurement) Error
Nonresponse, badly worded questions, interview error.