Armen, thank you very much for joining us in this this week on Sampling.

The topic that we're treating this week.

Could you please introduce yourself to our students?

Let them know who you are, where you work, what you do.

>> Yeah, my name is Armen Hakverdian, I'm a political scientist,

I mostly study public opinion and political representation.

So for instance whether political outcomes are influenced by public opinion or not.

The extent to which citizen preferences get translated into

policy are not very core issues of, of democratic governance.

That's what my research is about.

I also do a little bit of, research on political trust and

on the role of education in, in politics.

When it comes to teaching I mostly teach research methods.

>> Excellent.

>> so, at the bachelor level, but also at the, at the master's level.

>> Okay. >> Both, both statistics and

research design and, that sort of thing.

>> Yeah, wonderful.

I'm very to have you especially because your specialty is political sciences and

we've talked to sociologists, social psychologists, methodologists but

nobody from political sciences and I think you can give us interesting insight in

sampling because that's an important issue in in your field.

>> Yeah. >> So

could you give us a concrete example of how probability sampling is applied?

How, how complicated can it get?

>> It is incredibly complicated.

>> [LAUGH] >> It really is.

I mean, people underestimate how difficult it is to translate theory into practice.

It is extremely, extremely difficult, and time consuming, costly.

Because, I mean, the theory states that having a probability sample has

certain desirable characteristics.

It helps you to generalize to a particular population from which it was drawn,

that sample.

>> Mm-hm.

>> And and to, you know, it allows us to give estimates of where

a particular population parameter might be, with a degree of certainty.

>> Right.

>> That, that's incredibly elegant.

It's also very old.

I mean, it goes back to the twenties and

thirties, but getting that into practice is very difficult.

So, for instance, it starts with what the population actually is.

>> Hm. >> So,

usually when I, when I do research, and a lot of political scientists with me,

we don't want to generalize to the Dutch population as, as such.

I'm not interested in children, for instance.

>> Yeah. [LAUGH]

>> I mean, I'm interested in

the electorate.

The electorate is a subgroup of the population, so

we first have to specify what that population is.

I mean in the Dutch case, for

general elections, you would have to have dutch citizen, citizenship, er,

be eighteen years old or older, er, and there are certain

situations with mental incapacity where people are, are not allowed to vote.

This by the way is different from local elections where you don't have to

be a Dutch citizen.

So that's already different so if I want to, if I want to say something about Dutch

national elections, that is my population, the electorate is my population.

To front an electorate I need to have a list of everyone who's in it.

Because that's what I like to generalize to.

>> Yep, the sampling frame.

>> That is, yes.

I need to, and that is almost impossible.

The information is there, it's not publicly accessible.

So what for

instance in one important survey, the Dutch Parliamentary election study.

>> Mm-hm.

>> What happens there is that the we don't do it ourselves, but it's

handed over to a government agency, the [FOREIGN], the Central Statistical Agency.

>> Right. >> Who carries this out because they have-

>> Access. >> Access to, to those to those, to,

to that information.

Now, once you have the list of everyone who's in it, you have to,

to sa, you know, you have to draw the sample.

>> Mm-hm. >> You have to select respondents.

If you do this with, if you, you know,

if you use probability sampling then, usually the computer does this for you.

>> Mm-hm.

>> There's all sorts of different ways to sample it but

the computer usually would will, will draw a random sample.

But then of course you're still nowhere because you have to contact these people,

and that is incredibly difficult.

>> Mm-hm.

>> So sometimes you can you can phone them up.

You know, or, or open a web site and, and

have people, visit that web site and fill it in.

You can send interviewers to people's homes.

All those different modes will get you different results.

>> Hm.

>> So for instance, again, the Dutch parliamentary election study uses

interviewers, face-to-face interviewers.

>> Right.

>> But, not everyone who is invited to participate will actually participate.

>> Exactly.

>> So that response rate is, again,

it depends very much on the survey mode that you use, the type of survey.

>> You said, I know you get different responses, if you use different modes.

Do you mean different, response rates?

>> Yes. >> Or actually different answers?

>> Both.

>> Both response rates will differ and

both answers to questions will differ, so for instance a lot of

American pollsters still adhere to this principle of probability voting.

>> Mm-hm.

>> Using mostly telephone telephone sampling, or-

>> Probability sampling.

>> Yes, it's probability sampling, but

then they contact the respondents by phone number.

>> Okay, yeah.

>> Of course, throughout the history of polling this used to be using land lines,

land lines.

>> Hm. >> Telephone land lines.

And then they use random digit dialing.

So, the computer basically just picks a random number, that is a telephone number.

>> Tries it out sees if it's-

>> Yes, and then sometimes, you actually get a person on the other end.

But sometimes you get an automated.

>> Oh.

>> voice, robo-calls.

>> Yeah.

>> And then, but the response rates for

these types of surveys are not higher than 10%.

Wow, 10%.

>> Yeah.

>> Ooh, that's low.

>> So yes, that's incredibly low.

>> Yeah. >> And obviously,

the ones that participate are probably not a random subset of those-

>> Exactly.

>> Who you invited to participate.

So but with,

with face to face interviewers the response rates are much higher.

But it is incredibly expensive because you

have to pay trained interviewers to do the data collection for you.

So a survey like the Dutch Parliamentary election study,

NKO, in Dutch: Nationaal Kiezers Onderzoek.

Costs somewhere around half a million euros.

>> Whoa.

Okay.

>> And it is conducted over multiple months during each election.

>> Mm-hm.

>> Sometimes multiple waves, I mean often multiple waves,

so that adds to the cost, but it, it costs half a million Euros, or less.

>> So how many respondents take part?

>> A few thousand.

>> A few thousand?

>> Yes. >> That's it?

>> Yep. >> A half a million?

>> Yep. >> Wow.

>> Yep. That's expensive.

>> It's expensive.

But the, so but, the, the, I mean,

the benefit of it is of course, that you get a, you get a very high quality survey.

And much of what we know about Dutch voting behavior, come from the NKO.

So it is, it has a very long tradition.

iI's been collected since 1972.

>> Okay.

>> But obviously not everyone, almost no one has those sorts of-

>> That kind of funding, no.

>> That kind of funding.

So, yeah you have to find other ways.

You have to find other ways.

But, the fact of the matter is that the practice is very different from theory.

Even using the, the probability samples with,

such as, for instance, the NKO does, or the American pollster do,

you don't end up with a random sample the way that theory dictates.

>> Exactly. >> There's always biases in

there that have to be adjusted afterwards.

Sampling - Interview - Armén Hakhverdian (Part 1 of 4)

计量方法

与讲师见面