TU-L0022 - Statistical Research Methods D, Lecture, 2.11.2021-6.4.2022

Measurement is an important part of social science research. If we take a look at this research process diagram from Singleton and Straits we can see that the two most important decisions after you have decided on your research question is what do you sample, what are the units of analysis and what do you measure, which means what are the variables that you study from those units of analysis. After that you make your data collection and you do your data analysis and report the results.

The important part is that the quality of your study is mostly determined by what do you sample and what do you measure from your sample. So when you have your data then the upper limit of the quality of the study is basically determined. If your measurement doesn't work or if your sample is somehow flawed then no matter how complicated or how sophisticated analysis you apply to those poor data your research output will not be very high.

The idea of measurement is that we want to assign numbers to some quantities that we study. For example some things that we could study are heights of people, temperature outside, intelligence of a person, innovativeness of company. The idea of all these quantities is that they are variables. The idea that something is a variable means that it varies. Some people are taller than others. Sometimes it's colder outside. Sometimes it's warmer outside. Some people are smarter than others. Some companies are more innovative than others.

So the idea is that there's some kind of variation in the objects or the units that you study and the idea of measurement is that you want to assign some numbers to that variation, to quantify that variation. There are three key questions when you do measurement.

The first question is where do you get the numbers? So how do we assign the height of a person? How do we quantify it? So for height that's obvious we use a measurement tape for example. For temperature you use a thermometer but there are different kinds of thermometers that you can apply. But how do you quantify things that are not physical quantities like innovativeness or intelligence? That's a less straightforward to do and there are different ways of doing it.

The next question is what does the number tell you? So if you say that a company's innovativeness is 5. Is it a lot or a little? What does it actually mean? So we're talking about the meaning of the number and the interpretation of the number. '

Finally, how do we justify the way we assign the numbers? So we of course, besides just getting the numbers we have to convince our readers and ourselves that our numbers are actually valid for the purpose that we're using them for.

There are a couple of different, higher-level ways of getting the numbers. Let's look at the research designs by Singleton and Straits. They present four research designs. The first is a laboratory experiment. The idea of laboratory experiment is that you don't actually measure the key variable that you're studying instead you manipulate it. So laboratory experiments and experimental studies are more about manipulation of things than measurement of things. The remaining three are about measurement and they are different approaches of measurement and to some extent sampling as well.

The idea of a survey is that you measure things by asking people. So the subjects provide the numbers. If we study people, their intelligence, then we ask them whether they're smart or not. And if we study companies we ask people in those companies whether the companies are innovative or not. We can do it also indirectly by asking whether the companies have been successful in producing new products and new services.

And then we have, the second category of field research. The idea of field research is that we don't ask the subjects instead we rate or our research assistant rates or evaluates the subjects and records what happens based on observation and that gives us the numbers.

Finally we can use numbers collected by others so that's the archival records.

So that's basically how we get the numbers. The actual practicalities of how they do that is something that I'll address in other video but those are the three main main ways of getting the numbers. Ask the people, rate yourself, use data collected by somebody else.

The next question is what do the numbers tell us and how do we justify the numbers? To answer those questions we need to understand a little bit about measurement theory which relates to how the data and the thing being measured are related. To understand measurement theory we need to understand the concept of a latent variable. The idea of a latent variable and observed variable is that we have two types of variables. The observed variables are variables for which we have case values. So we have a specific number for each individual in our sample. We have a specific number for innovativeness of the first company, second company and so on.

These are in model path diagrams, these are presented by these squares. Sometimes measurement measured variables are called indicators or manifest variables which highlight that the purpose of these measured variables is oftentimes to quantify some unmeasurable or unobservable thing.

Latent variable is another kind of variable. The idea of latent variable is simply that it is a variable for which we don't have the case value. So we know that there is some variation between companies or between people but we cannot specifically assign numbers to any companies. We just know that there's some variation on some attribute or some variable but we cannot assign the exact numbers.

We can estimate these numbers and we can estimate correlations within latent variables so we can't say what the specific values are. So the difference between latent variable and observed variable are important when we talk about measurement theory and when we talk about models that allow us to test or use or operationalize our measurement theory.

Then we need to understand a couple of other terms as well. We have to understand the difference between concept, construct and measure.

The idea of a concept is that it's an abstract label for things that we study. And concepts have a reference and often the meaning as well. The idea of a reference is that for example if we have a concept of rock that refers to certain objects that we call rocks. The idea of meaning is that the concept has some kind of meaning as well. So if if we say that we have a rock then people know that well we have something that is naturally occurring. It is a hard object. It's probably size of a couple of fists or so. That's it instead of a boulder which is a larger one and so on.

So the term has some kind of attributes or the concept has some kind of attributes that are attached to it that give it meaning. Then a concept can also have a definition and the idea of a definition is that we have agreed on a specific written definition of what exactly the concept means. When you read papers that develop theory or introduce new constructs then they quite often define the construct explicitly. I'll get to constructs in a moment. Examples of concepts are persons for example and rocks and many other things. So concepts are like abstractions of things that we can observe and study.

A construct is a special kind of a concept. It is a conceptual variable. So the idea is that it can vary. So people or organizations can have different decrease of a construct. You can have a different degrees of innovativeness different amounts of intelligence and so on. So whereas in these concepts they can refer generally to just about anything, construct is something that is typically quantifiable. And because these are quantifiable we can study constructs using quantitative techniques.

Constructs are also latent, in the meaning that we cannot assign explicit correct values. We can only observe constructs indirectly. Constructs also can have dimensions for example we could have a construct of a person's size with the two main dimensions of height and weight of the person. Then some examples are intelligence, it could have some dimensions. Innovativeness, it could have dimensions. So for example how well you are doing in product innovation and how well you are doing a service or process innovation and so on.

Then measure is the third kind of variable or a thing that you need to understand. Measure is an observable variable that quantifies one dimension of a construct. If you have multiple dimensions in a construct then you need at least one measure for each. So it doesn't make any sense to try to quantify person's size using one number. You need at least two numbers: the height and weight. Examples include IQ test scores so that a measure for intelligence and reading on mercury column thermometer which is a measure of temperature.

How do these construction and measures then relate? There are two main approaches. One is a nominalism. The idea is that in nominalism you basically reject the existence of constructs independently of measurement. And an extreme version of nominalism is operationalism which says that the construct is simply whatever the measurement process produces. So the construct is defined by the measure.

And then realism assumes that the constructs exist independently of measurement and the purpose of measurement is to discover the true values of the construct. Most social science research follows the realist approach. So the idea is that there exists something called innovativeness independently of our measurement. We can say that some companies are more innovative than others without measuring those. So that kind of statements make sense if we assume that innovativeness or intelligence exist independently of our measurement attempts.

Then how we actually apply these concepts in practice is that we use the measures as proxies for the constructs. We cannot really observe the constructs directly so the next best best thing is that we build some kind of statistical presentation based on our data. And for example, we can just use a number as such, we can take a sum of multiple numbers or we can build a latent variable model and then use the latent variable as a proxy for the construct. So we use these empirical representations constructed based on our data as proxies for the constructs assuming that the empirical representation is a perfect representation of the construct. That is of course something that is hardly ever exactly correct but we have to justify that it is a good enough approximation. So that's the idea of a proxy. Instead of using the construct when we study something, we use the measure as a stand-in for the construct.

Summary of these key concepts. We have the constructs. Construct is a concept. So it's a variable that exists in principle. It can have some definition, or almost always has a definition. We can say that some companies or some individuals are higher on the construct than others and we cannot observe it directly. Then observed variables are specific numbers for each subject or a case that we have collected somehow. The idea is that these measures are - if we take the realist perspective - the idea is that the variation in these measures is caused by the variation in the construct. For example people's IQ scores differ because their intelligence differs. So the reason why there's variation in the data is that there's variation in the construct. Thermometer changes its value because the temperature outside is different from one day to another. So that's the idea of realist perspective to measurement which I will be using in these videos.