Conceptualizing Your Research

So here we are at class 4, Conceptualizing Your Research.We're going to start with an overview of the researchquestion.With survey research, the research questionis key and requires a lot of your attention,because your research question-- or research questions--lead to your survey questions.

So there's a logical sequence between the research questionand your survey questions that appear on your survey.The research question, as you've heard many timesbefore in this course, is a "clear" statementof what you want to know.The word clear is critical here.Research question should be concise and unambiguous.

So just to recap, with survey research,the research questions specify whatyou're trying to describe, explain, or influence.So how do we go about refining the research problem?This is when you act like a sponge,taking in as much information as possible.

You want to inform your research problem from every direction,so there are several ways of doing this.One is by conducting a literature review.The literature review will tell youthe state of art in the field in which you are interested.Literature review should be integrative and all-inclusive.Of course, in today's world, all-inclusive

can be somewhat problematic, because therecan be no end to what you can research and learn and know,so you need to put some parameters around your researchproblem.Another way of refining your research problemis to talk with experts, colleagues, and clients.These can be informal chats around the water cooler or more

formal conversations to inform yourselfabout your research problem.A third approach is to hold focus groupswith people similar to the ones you want to survey.The purpose of the focus group isto inform your research problem by getting the focus groupparticipants' input into the importance of the topic,

the relevance of the topic, and how it could possiblybe sharpened.So once you've defined your research questionsand sharpened your research problem,then we need to go to the next step.So let me start with an illustration.I'm going to put up some words in this bubble.

Social class, love, power.What are these words?How do you measure them?Can you measure them?These words represent concepts, and concepts are the buildingblocks of any study.Can we measure love?

Can we measure power?Can we measure social class?The answer is no and yes.We can't measure these concepts directly,but we can measure them indirectly.But the first stage of designing a surveyis to define the concepts in their broadest sense.

So what is a concept?It's not a single measurable entity.At this point, it's in the form of the abstract, whichwe're going to eventually going to turn into the concrete.So rather than measure concepts directly, weneed to measure them through other directly observablevariables, and these are the variables

that appear on our surveys.So let's start with an example.Say we're interested in beliefs.That's a very broad topic.So what do we do?We probably read at the very general level about beliefs,only to learn this is a massive topic which

requires a lot of narrowing.So how do we narrow?One way is to make it discipline-specific.So we could be interested in beliefsabout religion or beliefs about civil society, for example.But in this example, we're going to lookat beliefs about education.So just by doing that, we've narrowed the topic

down dramatically.But is there a single concept about beliefs about education?The answer is no, but how many concepts will webe working with?This is your decision, and it's the ideaof delimiting your research, where you narrow downthe topic to make it manageable for the purposes of your study.

In this example, I've chosen two beliefs.So they're both generally about education,but they're delimited to beliefs related to workand beliefs related to leisure.You'll get the education part in a minute.So concepts somewhat magically morph into what we call latent

constructs.Concepts are broad, they're at the conceptual or theoreticallevel, and as we move toward latent constructs, that'swhen we move into a measurement model.We're determining how we're going to actually measurethis concept, now a construct.

As I said, it's very subtle and just slightlya shift of a way of thinking.So we've just seen two concepts, which now we'velabeled as latent constructs, and the reason they'reconstructs is because now we're going to try to measure themempirically.And how do we do that?Well, we need some survey questions.

So on one hand, we've done a lot of conceptual workto define the concepts and constructsand now we need to apply that knowledgethat we've gained to formulating survey questions.This is the concrete part and this is critically important,because your constructs can only bemeasured by these very concrete measures you

have on your survey.And these are called indicator variables.This are the real live questions you have on your surveythat you're going to use to measure your constructs.So in this example, we have six.Take a moment to read them.

OK.Now we're back to conceptual work.We've developed constructs.Now, we've developed indicators to measure these constructs.Here, we've identified belief one, two, and threeto relate to construct number one, and belief number four,five, and six to relate to construct number two.

However, this might not necessarilybe the case when you analyze your data,because it may be that belief number one actually belongswith construct number two.And there could be considerable movement,but to the best of your ability, youconceptualized your constructs, and you've created indicators

to measure those constructs.And then there might be some surprises along the waywhen you do your analysis, and thisis an example of a surprise.So what we've just done is create hypotheses.I think hypotheses are best described as educated guesses,and we all make educated guesses all the time about everything,

except in this instance, we're doing itfrom an academic perspective.So an hypothesis is the act of linking key constructsto indicator variables.Here, we are making an educated guess.What goes with what?It is a guess, because we're not sure-- that'swhy we're doing the study-- except it's an educated guess,

so it's the best guess we can makegiven the knowledge at hand.That's why gathering knowledge beforehand is so important.An hypothesis is also a specific statementsabout a phenomenon that can be tested empirically,because that's what we're going to do.We're going to test our hypothesesusing indicator variables.

As an educated guess, our hypothesesstate what should happen in a particular situation.An hypothesis always involves at least two variables,and we need to identify the nature, direction,and strength of the relationship between these two variables.So you must specify each of these.

The nature, the direction, and strength of the relationship.What is the nature of the relationship?So what are the two variables under consideration?Two or more.Next is direction.In our example, do beliefs about worklead to beliefs about leisure, or is it the other way around?

And finally, the strength.Is there a strong, positive relationshipbetween beliefs about work and beliefs about education,or is it weak-- negative?So we use the words, weak, moderate, strong,negative, and positive to describethe strength of a relationship.

Latent constructs are abstract, and they're notdirectly measurable.When we draw them, they're always represented as a circle.Not an oval, not a square, not a rectangle, but a circle.This is really important, because thisis what the literature recognizes so wemust follow these protocols.

Indicator variables are the measurable indicatorsof our concepts.These are the tangible measures on our surveys--the real questions-- that we used to measure the latentconstructs.They are always represented in a model by a rectangle.There are a couple of other terms we must consider as well.

The first is dependent variable.The dependent variable is a key variablethat needs to be described, understood, or influenced.Another way to think about it is what needs to be explained.The next term to be explained is the independent variable.And these variables are those that mightaffect the dependent variable.

In other words, what things might do the explaining.So take a moment and think about your own research interestsand your own research questions and think aboutwhat needs to be explained, and what might do the explaining.Another type of variable to consideris a mediator variable.

So in everyday language, a mediator comes between.Think about a mediator in terms of a marriage breakdown.A mediator comes between the two partiesand tries to work out a solution to save the marriage.In terms of data analysis, a mediatorcomes between, or mediates, an independent variableand a dependent variable.

Here's an example.a is the independent variable, the variablethat might do the explaining.C is the dependent variable, the variablethat is to be explained.And b is the mediator variable.It comes between the two.So there might be a direct relationship between a and c--

that's the arrows say.Or there might be an indirect relationshipthat goes through b.So the relationship is from a to b to c.The b variable mediates the relationship between a and c.The last type of variable we would like to consideris a moderator variable.

Moderator variables are somewhat different,because they do not cause, lead to, or influencedirect or indirect effects.Instead, they're variables that can be subdividedinto different levels.Some examples include gender, age, or geographic location.So we could divide our sample into rural and urban.

The relationship between an independent variablea, and dependent variable b, may differ basedon the level of the moderator.So our results could be different for the rural samplein relation to the urban sample.So rather than having b mediate the relationship between a,and c, b is really a subdivision.

It's a subdivision of the sample thatmoderates the relationship between a and c.Your assignment in relation to this classis to draw a model for your study.And as I said, there's some protocols to be followed.The dependent variable is always represented by y.

The independent variable is always represented by x.If you have more than one independent variableyou identify them by numerical subscriptsand then you must indicate how x affects y.So we're back to our hypotheses, because our hypotheseswill tell us that.And because you've already identified your hypotheses,

it's just a matter of drawing arrows.It's also important to consider how y is affected by x.And just as a reminder, latent constructsare always represented by circles,and indicator variables are alwaysrepresented by rectangles.I literally copied these and I put them into my model,

so I'm consistent.So here's a little model.Variable a is beliefs about work.This is our independent variable.Variable c is choice of post-secondary program.So what the model indicates is that Iexpect there to be a relationship between beliefs

about work and choice of post-secondary program.The nature of the relationship is clear,the direction is clear.Beliefs about work leads to choiceof post-secondary program.There's no indication of strength here,but I would have specified that in my hypothesis.So let's say I expect a moderately strong relationship

between beliefs about work and choiceof post-secondary program.So beliefs about work and choice of post-secondary program areour constructs, which we cannot measure directly,but we have indicators.We have x1, x2 for beliefs about work,and y1 and y2 for choice of post-secondary program.

However, we also have a mediator variable.And this is beliefs about leisure.So this is a little chink in the study.So we have a nice clear relationshipbetween a and c but now we have a b.So what I've indicated is that beliefs about worklead to beliefs about leisure, which then lead to choice

of post-secondary program.So beliefs about leisure mediate the relationship.Let's say beliefs about work are measured by oneto help me prepare for a job and to increase my income,and this variable is measured on a four-point scale-- strongly

disagree to strongly agree.We'll get to that part later.And choice of post-secondary program.y1 could be non-university versus universityparticipation, and y2 could be choice of discipline.But then we have this mediator variable-- beliefsabout leisure.

Let's say x3 is I value my leisure time,and x4 is I want to achieve work-life balance.Again, both variables measured on a four-point scale--strongly disagree to strongly agree.So take a moment and think about that.

So it might be that people have strong beliefs about workpreparing them for a job or work leading to higher income,but then the mediators suggest that they alsovalue leisure time.So will this affect their choice of post-secondary program?We'll have to find out.

So that's pretty complex, but it evengets a little more complex, because wecan throw in some moderators.Will this relationship differ by gender and willit differ by age groups, for example?These are the moderator effects.To sum up, we started talking about research problems

and research questions, which ledto a discussion of concepts, and the importanceof strong conceptualization for a study.How these concepts are then developedinto latent constructs for which we need indicatorvariables to measure them?

We discussed hypotheses in relationto nature, direction, and strength of the relationship,and then we did some model-building.We looked at independent variables, dependent variables,mediator variables, and moderator variables.Now, you're set to draw your model.