Skip to main content
Search form
  • 00:02

    INSTRUCTOR: Welcome to class 7, Errors and Corrections.Today, we're going to carry on from the previous classon sampling to look at some of the pitfallsinvolved, and of course their corrections.There are four types of error, non-sampling errors,sampling errors, non-response errors, and measurement errors.

  • 00:26

    INSTRUCTOR [continued]: I like what Kish has to say about error.He makes three points.First, if the problem is small-- and we'll define problem soon--and it doesn't bias the study, then it'sjust best to ignore it.If the problem is large, and thismeans that the sampling frame is highly flawed,then you need to rethink your study.

  • 00:48

    INSTRUCTOR [continued]: In other words, you need to match the purpose of your studywith the sampling frame at hand.His third point is to correct the errorsin the list of the population.First we'll talk about non-sampling errorsand the four different types.

  • 01:08

    INSTRUCTOR [continued]: The first is coverage errors.And let's start with an example.Say you're a pollster wanting to test the waters in the upcomingelection.And you're going to do a telephone survey.What happens when people supporting one partydon't have landlines but only have cell phones,and your only access to a list is a telephone book?

  • 01:33

    INSTRUCTOR [continued]: Here you have a coverage problem.Because those people you want to contact aren't on the list.If you proceed with the survey, you'regoing to introduce bias into the study.So some rethinking is required.The next type of non-sampling error

  • 01:54

    INSTRUCTOR [continued]: are those who are ineligible to participate in your studyaccording to your inclusion rules.Let's look at another example.I've been following the high school graduatingclass of 1988.A few of the graduates were much older than a normal high schoolgraduate.In other words, they were adults whocame back to earn high school credentials.

  • 02:16

    INSTRUCTOR [continued]: So in my study, they were ineligible.Because they are ineligible, if I send them surveys,it increases the costs and reducesthe sample size of my study.So what to do?

  • 02:36

    INSTRUCTOR [continued]: Ideally, we would like to screen ineligibles outbefore the sampling begins.Sometimes this is not possible.And I think in the case of my study,it was not possible because I didn'tknow their ages in advance.If it's not possible, then oversamplingmight be necessary.And then you'll screen them out afterwards.

  • 02:56

    INSTRUCTOR [continued]: Then you can remove the ineligibles after the fact.A third problem is duplicate or multiplicity.This is most likely to happen whenyou're using several lists to comprise your sampling frame.The end result is having the same person on the listmore than once.For example, if you're comprising a list of teachers,

  • 03:20

    INSTRUCTOR [continued]: you might use a list from the school board,from the College of Teachers, and other sources thatcontain lists of teachers.And it's likely that an individualcan be on several lists at the same time.Solving this problem is relatively easy.You just need to eliminate the duplicates.

  • 03:41

    INSTRUCTOR [continued]: The next is clustering, and it's the opposite of duplication.So think of a household, an old fashioned householdwith one telephone, except four people live in the house.So people are clustered in the house.But there's only one listing in the form of a telephone number.

  • 04:04

    INSTRUCTOR [continued]: This can be a serious problem in terms of samplingand might require you to rethink your strategy.The next type of error is sampling error.And this is a mismatch between the sample and the populationto which you want to generalize.

  • 04:26

    INSTRUCTOR [continued]: There is always some type of mismatch,and your goal as a survey researcheris to minimize it as much as possible.Because sampling error can be calculated,it's considered a gold standard in terms of survey research.However, it has several limitations.This is a really important.

  • 04:46

    INSTRUCTOR [continued]: Sampling error can only be used if the study design isbased on a probability sample.Also, standard error can only be calculated on a single variableand not on the complete study.So when you read or hear about standard error in a poll,the standard error is only calculated on one question, not

  • 05:09

    INSTRUCTOR [continued]: the entire poll.The calculation of standard erroris based on the rules of probability.Every sample drawn from a populationwill not be exactly the same.The calculation of standard errorrequires drawing multiple random samples

  • 05:29

    INSTRUCTOR [continued]: from the same population and plottingthe mean of each sample.This plot is the mean of the sampling distributionor the true population mean.And then the standard error is calculated from this plot.There is a formula, but it's not important

  • 05:50

    INSTRUCTOR [continued]: that you learn the details now.It's much more important to know when and when notto use a standard error.A third type of error is non-response error,and this occurs when the individualsthat you have chosen for inclusion in your studydo not participate.

  • 06:11

    INSTRUCTOR [continued]: One type of non-response error is unit level response.This happens when an individual does not respondto any part of the survey.In other words, they don't respond at all.If unit non-response is systematic-- sayall the men in your study didn't respond-- thenyour study would be biased.

  • 06:32

    INSTRUCTOR [continued]: Even studies with high response ratesmight be affected by non-response biasif certain groups are less likely to participate.Take a moment and think back to the last classon sampling and particularly stratified sampling strategies.This is a way to deal with non-response bias.Non-contact non-response happens when

  • 06:54

    INSTRUCTOR [continued]: it's not possible to locate an individual.In other words, the survey is undeliverable.This occurs when the contact information you havefor the individual is invalid.There are ways of validating mail addresses, for example,through Canada 411, a database of Canadians.

  • 07:17

    INSTRUCTOR [continued]: Another form of non response is incapacity.And this means that some people simply cannot fill outyour survey.It might be due to illness, or perhaps the personcan't speak the language of your survey.Or in the case of an internet survey,it might be due to a lack of skillin terms of using the internet.

  • 07:38

    INSTRUCTOR [continued]: This form of non-response error isusually random and very small.And finally, non-response occurs because people simplywill not cooperate.Individuals have the right to refuse to participateand the right to refuse to participate at any timeduring the survey process.If this is systematic, then the results will be biased.

  • 08:00

    INSTRUCTOR [continued]: So you must consider the questions on your surveyvery carefully.The last type of error is measurement error.And this occurs when the survey items do not measurethe constructs of interest.This is a survey design problem, and itmeans creating a good survey with solid questions

  • 08:21

    INSTRUCTOR [continued]: that people can understand and answer.So to help you avoid measurement errors,revisit the topic of questionnaire design.To sum up, we've covered four typesof error, non-sampling errors, sampling errors,non-response errors, and measurement errors.

  • 08:45

    INSTRUCTOR [continued]: Errors will affect the integrity of your study.So it's important to take them very, very seriouslyand avoid them if you can.

Video Info

Series Name: Designing and Doing Survey Research

Episode: 6

Publisher: University of British Columbia

Publication Year: 2015

Video Type:Tutorial

Methods: Survey research, Sampling error, Measurement error, Sampling

Keywords: communication barriers; cooperation

Segment Info

Segment Num.: 1

Persons Discussed:

Events Discussed:



Professor Lesley Andres explains non-sampling errors, sampling errors, non-response errors, and measurement errors. She discusses how to take the different types into account during survey design and analysis.

Looks like you do not have access to this content.

Errors and Corrections

Professor Lesley Andres explains non-sampling errors, sampling errors, non-response errors, and measurement errors. She discusses how to take the different types into account during survey design and analysis.

Copy and paste the following HTML into your website