6 Building a Theory, Model, and Hypothesis

Author

Moses Rivera, Ph.D.

Published

October 5, 2025

6.1 Introduction

This course focuses primarily on the hypothetico-deductive approach for inquiry—which I defined in sec-epistemology_IWO—because that is the predominant approach in the current praxis of IWO Psychology research. Other approaches are very valid, especially the qualitative approaches described in sec-qual. However, for the sake of your immediate employability and acclimation to the current world of IWO psychology, this course will focus more on the hypothetico-deductive approach for quantitative inquiries.

In this chapter, our main goal is to define several of the most important terms that will help you create theories or talk about theories, and to create good hypotheses to test a theory. Along the way, we must define several other terms that describe the processes involved in creating a theory and/or creating a hypothesis.

6.2 What is a theory?

A theory is an idea or statement about the relationship(s) among two or more concepts (compare with: Jaccard & Jacoby, 2020). The idea and/or statement may or may not be accurate or reflect “reality”, but it is still a theory. Thus, Figure fig-simplest-theory is a diagram of the simplest structure for a theory.

Some scholars (e.g., Jaccard & Jacoby, 2020) prefer to distinguish between a theory versus a conceptual system, such that a conceptual system is a mental representation about the relationship(s) among two or more concepts, whereas a theory is an expressed statement (usually written or spoken) that represents a conceptual system. Such distinction isn’t necessary in this course, but I felt it was worth mentioning.

Figure 6.1: The simplest structure for a theory.

In the language of mathematics, the diagram in Figure fig-simplest-theory is known as a graph, which comes from an area of mathematics known as graph theory. Laypersons typically use the word graph to refer to any data-based diagram (e.g., a bar chart). However, in this book, the word graph will always refer to the meaning from graph theory (i.e., the graph-theoretic sense).

The concepts and relationships in a theory can be anything you specify, and you can draw multiple valid graphs to represent any given theory.

Also, a relationship in one theory may be a concept in another theory, and vice versa. It’s all based on the perspective of whoever is writing the theory or drawing a graph for it.

Importantly, in any graph that represents a theory, each concept and relationship must be labeled, otherwise the correct interpretation of the graph is ambiguous. For example, consider the graph in Figure fig-unlabeled_graph, in which the relationship betweens productivity and income is unlabeled.

Figure 6.2: This graph has an unspecified relationship between productivity and income.

The relationship between productivity and income could be anything: maybe productivity increases income, or productivity is positively correlated with income, or maybe productivity is uncorrelated with income, or perhaps productivity is negatively correlated with income. Obviously, the latter three theories seem to contradict each other, so we would need to do some empirical tests to figure out which one is true. Similarly, there are many theories in science, including IWO psychology, and it’s up to us to test them to figure out whether they are accurate reflections of what we perceive as reality.

Another word worth discussing now is model. Some scholars (e.g., Jaccard & Jacoby, 2020) consider models and theories as synonyms for each other. Fundamentally, I agree with that definition, but many (most?) IWO psychologists (and/or behavioral scientists in general) typically use the word model to refer to either a part of a theory or a statistically derived mathematical representation of a theory. In quantitative research methods, you will almost always be creating a statistical model, such as Equation eq-simple_model, which is a math-symbolic expression of the theory that productivity is correlated positively with income:

$\text{income}=\beta_0+\beta_1\text{productivity}+\varepsilon, \text{ and }\beta_1>0 \qquad(6.1)$

Thus, the same theory and/or model can be expressed via words, or via graphs, or via math-symbolic expressions. Equation eq-simple_model is a math-symbolic expression of this verbal expression: “Productivity is correlated positively with income”. Likewise, Figure fig-simple_model is an equivalent diagrammatic expression from graph theory.

Figure 6.3: A diagrammatic expression from graph theory, equivalent to the verbal expression: “Productivity is correlated positively with income”.

Again, a theory is an idea or statement about the relationship(s) among two or more concepts. Next, let’s define what we mean by concept.

6.3 From concepts to cónstructs and variables

The word concept is a generic word to refer to anything that is conceived in the mind. Concepts can refer to things that have tangible manifestations (e.g., the concept of a worker’s height) or they can refer to things that are not so clearly tangible, if at all (e.g., the concept of being satisfied with one’s career). Also, concepts can be highly specific with rich detail (e.g., the concept of my uncle Steve’s productivity in December), or they can be more generic with less detail (e.g., the concept of any worker’s productivity in general). When we strip away the rich detail of a concept to make it more generic, we are moving toward greater abstraction. In other words, a more generic concept is more abstract than a highly-specific concept.

Recall from sec-falsificationism I described that scientists are generally interested in discovering universal statements that are true. In other words, IWO psychologists typically aren’t interested in knowing whether my uncle Steve’s productivity in December is correlated positively with his income. Rather, we want to know whether the productivity of all workers at Company ABC is correlated positively with their income, or perhaps whether that’s true for all workers in the USA, or perhaps the entire world! Thus, IWO psychologists are typically interested in concepts that are more abstract, rather than less abstract.

Abstract concepts are thus so special that scholars typically call them cónstructs (more typically spelled: constructs), rather than calling them concepts. Thus, a cónstruct is merely a generic term to refer to any concept that is regarded as being more abstract than not. There is no special rule to determine the level of abstraction that is appropriate before you can call a concept a cónstruct. For example, the role of engineer may be considered a concept, and so is the role of marketer. Thus, we could say the concept of roles in general is a cónstruct. However, we could just as easily say the engineering role of Steve is a concept, and so is the engineering role of Alex, and that the role of any engineer in general is a cónstruct. Thus, the distinction between a concept versus a cónstruct is in the eye of the beholder. In general, IWO psychologists typically refer to most concepts they study as cónstructs, perhaps out of habit.

Cónstruct vs. Construct…

The word construct can be a verb (i.e., to build)—in which case it is pronounced /kənˈstrək(t)/, like con-STRUCT—and it can be a noun (i.e., an idea or theory)—in which case it is pronounced /ˈkänˌstrək(t)/, like KAHN-struct. There are few instances where the meaning isn’t disambiguated via context clues. However, I always force the disambiguation via the acute accent diacritic over the letter o: cónstruct. As far as I know, there are very few people who write it like that, but I think it helps make things at least a little easier for new students to understand sentences like: “we’re going to construct a measure of a construct” (i.e., “we’re going to construct a measure of a cónstruct”). For me, the disambiguation is worth the extra keyboard strokes. Don’t be alarmed if you never see anyone write it like that after this course (and I won’t be offended if you prefer to just write construct!).

A special type of cónstruct that is commonly studied in science is called a variable. In the general vocabulary of science and data-analysis, a variable (pronounced: /ˈverēəb(ə)l/, like VERY-uhbl) is any cónstruct that represents multiple possibilities. For example, the temperature in any room of a company’s building is a variable, because there are multiple theoretical possibilities that the temperature could be (i.e., it could be cold or warm today, and some rooms might be different than others). Once we measure a room and we know its temperature is 70℉, then we say 70℉ is that room’s datum or value or score on the temperature variable. There are countless (arguably infinite) variables we can think of about people: variables about their appearance (e.g., the presence of tattoos, the person’s height), variables about their behaviors (e.g., how fast they talk, how nervous they typically feel), variables about the structure of their social networks (e.g., the types of coworkers they have), and the list goes on and on. Indeed, the cónstructs that are typically studied in IWO Psychology are almost always variables. Keep in mind: variables don’t have to represent numbers. A person’s hair color is a variable (i.e., some people have black hair, others have brown, blonde, red, etc.), but it isn’t necessarily represented by a number. However, because many IWO psychologists focus on quantitative methods, the variables they study are either inherently numeric or they are converted into numbers (e.g., black hair = 1, brown hair = 2, …).

Now that we’ve defined concepts, cónstructs, and variables, let’s now explore a special type of variable known as a dimension.

6.4 Dimensions and multidimensionality

You’ve likely heard that we exist in three-dimensional space, or that a drawing on a flat sheet of paper is approximately two-dimensional. We can use our intuition from those examples to arrive at a general definition for the word dimension. A dimension is any variable whose data can be put into a logically ordered sequence of increasing (or decreasing) intensity, amount, magnitude, or level of the variable’s essence. Common examples of dimensions are length, weight, color saturation, sweetness (as a taste), and size. Each of those is a dimension, because they are variables whose data can be put into a logically ordered sequence of increasing amount. For example, I could describe three foods along the sweetness dimension as not sweet, a little sweet, and very sweet. As you can see, as we move from left to right, those data are strictly increasing in their level of sweetness in that order—and they’re strictly decreasing in sweetness if we move from right to left. In contrast, a variable such as occupational industry sector with data such as agricultural jobs, healthcare jobs, and education jobs, is not a dimension, because those data do not form a linear pattern of any increasing (or decreasing) essence.

Thus, it’s convenient to think of a dimension as an imaginary directional line, such that each subsequent datum has more of the variable’s essence as you move in one direction of the line, and they have less of the variable’s essence as you move in the opposite direction of the line.

Datum is the singular word for data. For example, if you are doing a survey of everyone’s age at the university, and only one person responds to your survey, you have a datum. If multiple people respond to your survey, you have data.

Whereas dimensions like length, weight, color saturation, and size can be measured using tools such as rulers, weight-scales, and optical sensors, sweetness is a subjective perception. Instead of measuring sweetness, we could measure amount of sugar with a chemical device, but what if we are actually interested in the perceived sweetness of a food, rather than its amount of sugar? For example, diet soda drinks have no sugar, yet we perceive them as sweet and we may legitimately want to compare the perceived sweetness of several non-diet and diet drinks, regardless of whether or not they contain any sugar. To measure the sweetness perceived by a person, we could either ask them to tell us how much sweetness they perceive, or we could measure the parts of their brain that light up when they taste something sweet.

Similarly, in IWO Psychology we’re often interested in variables such as happiness, anxiety, or extroversion—none of which can be directly measured with a ruler nor weight-scale. That is why psychologists invented techniques to help measure subjectively perceived things and other abstract psychological cónstructs. We call those techniques psychometric techniques, and the topic of measuring such psychological cónstructs is known as psychometrics. In sec-chapter_sampling_psychometrics, I provide a gentle introduction to the basic ideas of psychometrics, but you will also receive an entire course on psychometrics later in this degree program.

In a nutshell, when we interview or observe people because we want to figure out how happy, anxious, or extroverted they are, we can use psychometrics to convert that information into numbers and that’s how we derive numeric variables for those cónstructs. Thus, for example, we can imagine a person’s level of extroversion can be represented by a number on an imaginary number line, starting from zero and increasing to greater levels of extroversion. Thus, we can think of extroversion, happiness, and anxiety as three separate dimensions. Of course, there are many more (arguably infinite) types of dimensions we could describe any person with. For example, each of these is a dimension:

The number of years the person has been working at their job.
The number of coworkers the person considers as their close friends.
The level of stress the person subjectively experiences on a typical day at work (e.g., on a scale from 1 to 10).
The level of passion the person feels toward their work (e.g., on a scale from 1 to 10).
The ease with which the person responds aggressively toward hostile comments from others (e.g., on a scale from 1 to 10).

As you can see, any person can be described via a never-ending list of dimensions.

6.4.1 Multidimensionality

When a variable is defined as a single dimension, we say it is unidimensional.

Sometimes, we may want to define a cónstruct as a combination of multiple concepts of multiple cónstructs (i.e., multiple ingredients). When two or more of those ingredients are distinct dimensions, then we say the overall cónstruct is multidimensional.

For example, the owners of Company ABC might define their worker’s job performance as a combination of the worker’s speed and the worker’s accuracy. Speed is a dimension, and accuracy is a separate dimension. A worker could be fast and accurate, or fast and inaccurate, or slow and inaccurate, or slow and accurate, or anywhere else in between! Thus, that particular definition of job performance is a multidimensional cónstruct.

In IWO Psychology there are many popular multidimensional cónstructs, such as:

personality (typically defined as a combination of five major dimensions: openness to experience, conscientiousness, extroversion, agreeableness, emotional stability);
job-satisfaction (typically a combination of one’s level of satisfaction with multiple distinct aspects of the job, such as one’s pay, one’s supervisors, one’s coworkers, etc.);
burnout (a combination of one’s level of emotional exhaustion, one’s amount of cynicism, and one’s level of reduced professional efficacy);
passion for one’s occupation (typically a combination of how much one likes the occupation, how much one is motivated to engage in the occupation, and how much one self-identifies with the occupation);
intelligence (typically defined as a combination of several dimensions of cognitive abilities such as verbal reasoning ability, and quantitative reasoning ability).

6.5 Types of relationships in a theory

At the beginning of this chapter, I said a theory is an idea or statement about the relationship(s) among two or more concepts. So far in this chapter, I’ve discussed the definitions of many types of concepts, including cónstructs, variables, and dimensions. Now let’s move on to exploring some important types of the other main ingredient in theories: the relationships!

Fundamentally, when you write a theory, the relationships you write can be whatever you believe is true. Your theory might or might not reflect reality, but that’s a separate issue. Just like we can define any concept in broad or specific detail, we can also define any relationship in broad or specific detail.

Giving examples is perhaps the best way to explain what we mean by a relationship between concepts. Let’s consider the examples in Table tbl-conceptual_relationship_examples, and notice it is split into Table tbl-deterministic and Table tbl-probabilistic. We’ll use those examples to explore some important types of relationships:

deterministic versus probabilistic relationships;
causal versus correlational relationships;
moderated relationships;
mediated relationships.

Table 6.1: Examples of types of relationships in any theory

(a) Deterministic relationships

Relationship	Hypothetical Example
`occurs before` or `occurs after`	“After working on a repetitive task for 10 minutes, boredom occurs”
`increases together with` or `decreases together with`	“The size of a worker’s house increases as their number of clients increases”
`increases as the other thing decreases`	“A worker’s risk of illness decreases as their salary increases”
`causes` or `is caused by`	“Disrespect within a team causes a decrease in psychological safety”
`moderates` or `is moderated by`	“The relationship between your skill-level and your performance depends on your motivation (i.e., the relationship is moderated by your motivation). In other words, your motivation moderates the relationship between your skill-level and your performance.”
`mediates` or `is mediated by`	“If you have more clients, you’ll have more income, which means you’ll go on more expensive vacations. Thus, the relationship between having more clients and expensive vacations is mediated by income (i.e., income mediates the relationship).”

(b) Probabilistic relationships (non-deterministic relationships)

Relationship	Hypothetical Example
`typically occurs before` or `typically occurs after`	“After working on a repetitive task for 10 minutes, boredom typically* occurs”*
`typically increases together with` or `typically decreases together with`	“The size of a worker’s house typically* increases as their number of clients increases”*
`typically increases as the other thing decreases`	“A worker’s risk of illness typically* decreases as their salary increases”*
`typically causes` or `is typically caused by`	“Disrespect within a team typically* causes a decrease in psychological safety”*
`typically moderates` or `is typically moderated by`	“The relationship between your skill-level and your performance typically depends on your motivation (i.e., the relationship is typically moderated by your motivation). In other words, your motivation typically moderates the relationship between your skill-level and your performance.”
`typically mediates` or `is typically mediated by`	“If you have more clients, you’ll typically have more income, which means you’ll typically go on more expensive vacations. Thus, the relationship between having more clients and expensive vacations is typically mediated by income (i.e., income typically mediates the relationship).”

6.5.1 Deterministic versus probabilistic relationships

As you can see, Table tbl-deterministic shows only relationships with deterministic phrasing, whereas Table tbl-probabilistic shows only relationships with probabilistic phrasing (also known as non-deterministic). If something is deterministic, it means it is guaranteed. If I throw a regular ball up into the air, it is guaranteed to come down. However, almost all relationships among concepts in psychology (and all behavioral sciences) are probabilistic—in other words, we can’t typically guarantee that we know what will happen, but we can assign some level of (un)certainty to our statement (i.e., we can assign a probability to our statement). Thus, in reading Table tbl-conceptual_relationship_examples, if you felt Table tbl-probabilistic is more reasonable than Table tbl-deterministic, then your instincts served you well!

6.5.2 Causal versus correlational relationships

You likely already have a great intuitive sense of what a causal relationship is. Intuitively, we understand that the sun causes the rooster to crow, and that the rooster’s crow doesn’t cause the sun to rise. In other words, the crow “listens to” or “takes orders from” the sun, but not the other way around.

In contrast, a correlational relationship doesn’t necessarily involve causation. In other words, some correlational relationships involve causation, but some don’t. Correlation is almost always defined numerically (which you will learn much more about in the statistics course of this degree program). For example, if I write down the time of day the sun rises, and also the time of day the rooster crows, and I do that every day for a week, I will see the sun rises and the rooster crows at around the same time as each other. Those two variables are correlated with each other. In this case, there also happens to be a causal relationship between them, because we know the sun causes the rooster to crow.

However, let’s look at another example where two variables are correlated but neither causes the other. Consider a situation in which several companies in the USA keep track of how many internship applications they receive each day, and they keep track for an entire year. Looking back on the year’s data, they notice the number of internship applications on any given day is correlated positively with the outdoor temperature of that day. In other words, hotter outdoor temperatures are associated with more internship applications, as depicted in Figure fig-correlation. Notice the arrow for the correlational relationship is bidirectional, meaning it has an arrow head on both ends, because correlation doesn’t imply any causal directionality. All we can say is that the two variables are correlated positively with each other.

Figure 6.4: Outdoor temperatures and internship applications are correlated positively with each other, because they increase together. We’re not saying either causes the other.

We know it would be ridiculous for the applications to cause cause the outdoor temperatures to go up, and it’s unlikely the outdoor temperature is causing a chemical reaction in the applicants’ brains that’s causing them to want to apply to internships. Instead, the true reason behind the positive correlation is that the applicants are predominantly college students who mostly only have time for internships during their summer break, which happens to be the hottest part of the year. Thus, the onset of summer causes the applications to increase and it also causes the temperature to go up, as depicted in Figure fig-common_cause.

Figure 6.5: Outdoor temperatures and internship applications are correlated positively with each other, because Summer causes them both to increase.

In almost all scientific theories, the relationships have directionality, which is why we use directional arrows to depict relationships in our graphs. When we are expressing a correlational relationship, we use a bidirectional arrow as in Figure fig-correlation. When we are expressing a causal relationship, we use unidirectional arrows as in Figure fig-common_cause. In causal relationships, the concept that the arrow is coming out of is called a predictor (e.g., predictor variable), also known as an antecedent (e.g., antecedent variable). The concept that the arrow is pointing into is typically called an outcome (e.g., outcome variable), also known as a dependent variable. For example, in Figure fig-common_cause, Summer is a predictor, whereas outdoor temperatures and internship applications are both outcomes.

6.5.3 Moderated relationships

A moderated relationship (also known as a conditional relationship or a modified relationship) is any relationship (between two concepts in a theory) whose characteristic(s) can change, depending on one (or more) concepts. For example, consider the graph in Figure fig-without_moderation which corresponds to this theory: “How much time I have available will influence how much time I spend with that person”.

Figure 6.6: A simple theory without any moderated relationships.

That seems like a perfectly reasonable theory, but I think we can all relate to the fact that we’ll probably end a conversation more quickly (or even avoid it altogether) if we don’t enjoy that person. In other words, we could say: “How much time I have available will influence how much time I spend with that person, but that depends on how much I enjoy that person”. This fancier theory is depicted in Figure fig-moderation.

Figure 6.7: A basic theory with a moderated relationship.

In Figure fig-moderation, how much I enjoy that person is a moderator (also known as a moderating predictor), and it is moderating the relationship between how much time I have available and how much time I spend with that person. We can also use any of these equivalent verbal forms:

The relationship between X and Y is moderated by M.
The relationship between X and Y is modified by M.
The relationship between X and Y depends on M.
The relationship between X and Y is conditional on M.
There is an interaction effect between X and M on Y.
There is an interaction between X and M on Y.
There is a two-way interaction between X and M on Y.

You could also specify a theory in which a relationship depends on two, three, four, or as many concepts as you like. For example: “The relationship between X and Y is jointly moderated by M and W”. In that case, it would be a three-way interaction effect, because X, M, and W are all interacting with each other in their relationship with Y.

Another fun fact about moderated relationships is that there are always at least two equivalent ways to draw a graph for a moderation relationship. For example, in Figure fig-moderation_equivalent, the position of how much time I have available and how much time I enjoy that person are swapped, compared to their positions from Figure fig-moderation. Thus, Figure fig-moderation_equivalent corresponds to this verbal expression: “How much I enjoy that person will influence how much time I spend with that person, but that depends on how much time I have available”. As you can see, it is conceptually identical to the theory expressed by Figure fig-moderation.

Figure 6.8: The theory expressed in this graph is identical to that of Figure fig-moderation.

Indeed, in any moderated relationship as the one depicted in Figure fig-moderation_generic_equivalent, it doesn’t matter which of the interacting predictors are thought of as the “focal predictor” versus the “moderating predictor”. The decision is based purely on how you prefer to think about your theory. Thus, Figure fig-moderation_generic_equivalent-a is conceptually identical to Figure fig-moderation_generic_equivalent-b.

(a) We think of X as the focal predictor, and we think of M as the moderator.

(b) We think of M as the focal predictor, and we think of X as the moderator.

Figure 6.9: Two equivalent expressions for the same interaction effect between the predictors X and M on the outcome Y.

6.5.4 Mediated relationships

Consider this theory from Table tbl-probabilistic: “If you have more clients, you’ll typically have more income, which means you’ll typically go on more expensive vacations”—expressed as a graph in Figure fig-mediation_example.

Figure 6.10: A simple mediation model.

Figure 6.11: A simple mediation model.

In that theory, the relationship between how many clients you have and how many expensive vacations you can go on is mediated by how much income you have (i.e., how much income you have mediates the relationship). Thus, we would say Figure fig-mediation_example is a mediation model, or a theoretical model about mediation.

A mediated relationship (also known as an indirect relationship) is any relationship (between two concepts in a theory) whose effect is transmitted through at least one intermediating concept (typically called the mediating variable, or the mediator).

These are equivalent verbal expressions about a mediated relationship:

The relationship between X and Y is mediated by W.
W mediates the relationship between X and Y.
There is an indirect relationship between X and Y, through W.

In this context, we’re not talking about the other sense of mediation which you might be familiar with from every-day English, in which a person serves as a mediator between two opposing parties in a dispute.

It’s also possible to specify a mediation model like the one in Figure fig-mediation_example_2, in which X has a direct effect on Y, while also having an indirect effect on Y via W.

Figure 6.12: A slightly more complicated mediation model.

6.5.5 Combining moderated and mediated relationships

Consider the theory expressed in Figure fig-conditional-process, which includes moderation and mediation—these are often called conditional-process theories (Hayes, 2022), as well as other names (see: Edwards & Lambert, 2007).

Figure 6.13: A conditional-process theory.

We could make it even more complicated by adding more predictors, including more mediators and more moderators anywhere in the graph, as well as more outcome variables. We might want to do that because we want the theory to more accurately reflect real life, which is often complicated. However, this brings us to a very important principle when we’re building a theoretical model:

“As the model matches reality better, it necessarily becomes less simple. Or, as it becomes simpler, it necessarily loses some of its match to reality” (Rodgers, 2010, p. 5).

Thus, a good scientific theory strikes a good balance between matching real life close enough to be useful, but not being so complicated that it becomes less useful because it only applies to a few specific situations or it is too complicated to understand.

An analogy…

Here is a good analogy for why the best theoretical models strike a balance between simplicity and complexity:

A map is a miniature model of the world. We can add more details to the map to make it match the real world better. However, if we add too many details to the map, the map starts to become too big to handle. Imagine if the map included every pebble on every road, every grain of sand, every blade of grass. If we tried to print that map on paper, it would end up being near the size of the entire world itself! If it were stored digitally on a computer, it would require an unfeasible amount of digital storage space and computer processing power. Thus, a good map is one that is a close enough representation of the real world to suit our needs.

This principle segues nicely into the next section: what else makes a good scientific theory?

6.6 Necessary versus optional features of a good scientific theory

At minimum, a scientific theory must be:

empirically testable (e.g., empirically falsifiable);
logically internally consistent (i.e., it doesn’t contradict itself).

The following features are typically desirable in a scientific theory, though not necessarily required:

increases our understanding of the world;
novel (i.e., it’s something new);
relatively broad in scope;
easily understood and communicated to others;
appropriately parsimonious (i.e., Occam’s razor);
consistent with other accepted theories that have achieved consensus;
accepted by the broader scientific community;
stimulates research;
addresses a real-world problem.

6.7 Finally: Hypotheses

Some scholars believe a hypothesis (plural: hypotheses) is fundamentally no different than a theory. In some sense, I agree with that. But I also know most scholars—especially in the behavioral sciences—tend to use the word hypothesis to connote the meaning from APA’s online Dictionary of Psychology:

hypothesis: an empirically testable proposition about some fact, behavior, relationship, or the like, usually based on theory, that states an expected outcome resulting from specific conditions or assumptions.

As you can see, that definition of hypothesis is consistent with our definition of a theory that is empirically testable. Thus, in the parlance of behavioral scientists, one of the most important features of a hypothesis is that it is empirically testable.

For example…

Here’s an example to help reify the distinction between a hypothesis versus a theory. Consider this theory: “An increase in an employee’s happpiness will tyically cause an increase in their performance.” Now, suppose I don’t yet have the resources to design a study that can test the causal mechanism, but I can design a self-reported questionnaire study that can measure whether employee happiness is correlated positively with employee performance. In that case, my hypothesis might be: “Self-reported employee happiness will be correlated positively with self-reported employee performance.” As you can see, that hypothesis is more clearly empirically testable than the original theory statement. If you feel like this hypothesis could be written with even greater specificity, you’re right! That is a topic in sec-chapter_sampling_psychometrics.

Typically, after you describe your theory using words, equations, and/or graphs, you would write one or more hypotheses that can be empirically tested. Then—assuming your hypothese are logically derived from your theory—when you or anyone else conducts an empirical study to test your hypotheses, that process accumulates evidence about whether your theory is true or not.

Research Questions vs. Hypotheses

Sometimes when we’re conducting an empirical study, we might seek to collect evidence to help us answer some questions that aren’t necessarily tied to any hypotheses. In that case, we might write one or more research questions (regardless of whether or not we also have any hypotheses in our study) that we seek to address via an empirical study. A research question is exactly what it sounds like: it’s a question in which you’re not making a statement nor prediction about what results you expect.

6.8 Suggested Readings

If you enjoyed this chapter, you’ll likely enjoy Jaccard and Jacoby’s (2020) book: Theory Construction and Model-Building Skills: A Practical Guide for Social Scientists. It is part of the The Guilford Press’ Methodology in the Social Sciences series, which contains excellent books about research methods.

I can’t help but share this funny bit of trivia: James goes by “Jim” and his middle name is Jay. Jacob (1940–2018) went by “Jack”. Thus, the authors are James “Jim” Jay Jaccard and Jacob “Jack” Jacoby. It doesn’t get better than that!

McGuire (1997) is a classic article with excellent tips to help inspire your creativity for creating a hypothesis.

References

Edwards, J. R., & Lambert, L. S. (2007). Methods for integrating moderation and mediation: A general analytical framework using moderated path analysis. Psychological Methods, 12(1), 1–22. https://doi.org/10.1037/1082-989X.12.1.1

Hayes, A. F. (2022). Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach (Third edition). The Guilford Press. https://www.guilford.com/books/Introduction-to-Mediation-Moderation-and-Conditional-Process-Analysis/Andrew-Hayes/9781462549030

Jaccard, J. J., & Jacoby, J. (2020). Theory Construction and Model-Building Skills: A Practical Guide for Social Scientists (Second edition). The Guilford Press. https://www.guilford.com/books/Theory-Construction-and-Model-Building-Skills/Jaccard-Jacoby/9781462542437

McGuire, W. J. (1997). Creative hypothesis generating in psychology: Some useful heuristics. Annual Review of Psychology, 48(1), 1–30. https://doi.org/10.1146/annurev.psych.48.1.1

Rodgers, J. L. (2010). The epistemology of mathematical and statistical modeling: A quiet methodological revolution. American Psychologist, 65(1), 1–12. https://doi.org/10.1037/a0018326