Help EDR610 : The Class : Variables/Hypotheses : Hypotheses : Lesson3-2-1

 Understanding Hypotheses

Just to give us an indication of "where we've been & where we're headed ... !"
  • We've seen that the "heart & soul" of research is the research question/problem statement;
  • And in addition, last time we took a look at some key components of these research questions/problem statements: namely, variables (the "things" of interest to us as combined in the question/statement).
  • This time, we'll look at a third dimension of the above; namely, our "guess-timate," if possible, as to what the possible 'answer to the research question' might be. Such statements are called hypotheses!!

    Or to put the above into graphical perspective:

  • Basic definition: A hypothesis is a prediction about the outcome of a study. This prediction is what we believe will "hold up" or be true for the population at large, as a result of what happens in our study sample. (Remember that we usually can't study everyone in our target population for practical reasons of time and cost, so we are faced with drawing a smaller sample for our study and then 'projecting' the sample results with some degree of confidence back to the target population at large.)

  • Therefore, since it's a prediction, it's stated in sentence form. This is contrast to the two alternative & acceptable forms for the 'curiosity:'
    • problem statement: sentence form; OR
    • research question: question form.
  • It will also focus on the key variables contained in the research question.

Example of a problem statement and its related hypothesis:

This study is to determine the effects of a peer-assisted method of teaching reading, as compared to the traditional method, in terms of reading comprehension.
Problem Statement: (Review challenge! Which family does the above belong to? -- > keyword "effects" would imply a more-or-less controlled setting, with the "new" method as the "treatment" and the traditional method as the "control." Right: experimental!)
Hypothesis: Students taught by the peer-assisted method of teaching reading will score significantly higher on a reading comprehension test than students taught by the traditional method.

Key difference between the above two forms (research question/problem statement & its related hypothesis):

The research question or problem statement is in "open-ended" form, while the hypothesis states a definite outcome or set of outcomes that we might predict!

This goes along with the definitions of the two!

  • The question/statement is the 'curiosity,' so while you may (& probably do) have some idea as to how things turn out, you are leaving it open to 'see what happens!'

  • In contrast, the hypothesis is the place to state your prediction, or guess-timate, of what will occur. Thus, it takes a definite direction, or lack thereof (e.g., if you'd predicted "no difference" between the two methods) -- in this case, your speculation (which could be based on your own or others' experience; a review of the literature on related research; or even hunches which are A-OK!) as to 'which way it'll go!' In this case the researcher is "hypothesizing" that the peer-assisted method of instruction will yield higher reading comprehension scores than the traditional method.

Now, you might think it's kind of "cheating," or "stacking the deck," to "go out on a limb" like this and state your prediction ... BUT that leads us into the one key property that ALL hypotheses should possess: namely, they must be in testable form!

For that will be the purpose of your research design and analysis:

  • to scientifically and objectively gather your evidence (data); AND
  • then revisit your original "guess-timate," or hypothesis and then decide:

    1. Shall I "keep believing it?" -- e.g., does your prediction "hold up" in the face of the evidence, or data? If so, you are "retaining" your hypothesis! OR
    2. Does the evidence/data you've collected from your study point in a different direction than the prediction you originally made? Is this evidence 'strong enough' to lead you, then to 'quit believing' your hypothesis and switch to another belief? This is called "rejecting" your hypothesis!

    ***: And, stats fans ... we'll soon learn how to make this determination quantitatively! That is: when is the numeric evidence from your study sample "strong enough" to "safely" reject your original hypothesis in favor of another belief about reality!!!

What makes an hypothesis "non-testable?"

Let's zero in on this all-important property of "testability" by "back-dooring" the issue -- that is, looking at what might make a hypothesis "non-testable!

These tend to come in two (2) forms:

  1. The researcher has accidentally left out a basis for comparison. Suppose, for instance, that he/she formulates the following hypothesis:

    "Students taught by Method A will be better readers."

    How can we really 'test' this hypothesis if we don't know for sure what 'better' means here?! "Better" implies a comparison, but we don't know what's being compared!

    • Is it Method A to some other method? if so, what's the other method?
    • Or is it the students to themselves: e.g., students before getting Teaching Method A, as compared to the same students after they've gotten Teaching Method A?!

    Note how, in our example of a peer-assisted statement vs. traditional instruction, the comparison is explicitly stated within that hypothesis. That is: Method B (peer-assisted) is being compared to Method A (traditional).

  2. Another form of nontestable hypothesis is one that is really more of the researcher's value judgment, or his/her opinion .

    Quite often, though, with a little work and effort (ah ... much like most things in life ... !!!), many such 'value judgement/opinion' statements can be converted into (testable) hypotheses!

    Example: "All junior high school age boys should be required to take a course in home economics."

    This is more of a value judgment or opinion! (Please note the "should"!)

    It may represent what you honestly feel or believe ... but from the way you've stated it, there's no way to "objectively & scientifically" put your beliefs to the test via a research design & analysis like our overall paradigm in Module #1!

    But one way to begin to 'massage' this into an eventually researchable and testable hypothesis might be to ask the person making the statement, "Why do you feel this way?"

    Suppose he/she replies: "Well, such a course would probably help these boys think & behave in less sex-role-stereotypic ways."

    Ah ... now we're onto something ... !

    Now, we will not be able to 'scientifically and objectively conclude' (e.g., on the basis of a sound research study design & analysis) whether or not it is "good" for boys to behave in less sex-role-stereotypic ways."

    BUT we CAN empirically test the question as to whether a course in home economics, say, will reduce such behavior.

    We might then propose the following hypothesis:

    "Junior high school age boys who have taken a course in home economics will exhibit significantly less sex-role-stereotyping behavior than junior high school age boys who have not taken a course in home economics."

    Please note the switch in focus! From opinions/beliefs to observable actions/behaviors!

    Granted ... the 'burden' on the researcher will be to precisely define such key variables as "sex-role-stereotyping behaviors" -- in a way that's considered 'acceptable' as per, say, prior definitions of what does or doesn't constitute such behavior.

    This is called "operationally defining your variables:" in a way that's generally acceptable and can therefore serve to determine whether that variable 'is' or 'isn't there' in a given situation. In our situation, we'll be able to 'objectively' look at a subject's behavior -- apply the operational definition -- & then determine, "Was or wasn't that particular behavior an instance of 'sex-role-stereotyping behavior' as I the researcher have defined it for my study?!'

    But once the researcher does this, he/she has 'objectified' the issue! There is a way to 'tell' or 'determine' if the behaviors decrease after the exposure of the boys to the home economics course. Therefore, this particular hypothesis can be tested -- and then, of course, retained or rejected once the study evidence is in!

    It's now out of the realm of opinion, prior belief, feelings, and into observable behaviors/actions -- e.g., it is now scientifically & objectively testable!

Now ... for some add'l. terminology regarding hypotheses that you may have already encountered!

Specifically, we'll look at the terms "null hypothesis" and "alternative hypothesis."

In our example of testing the peer-assisted method of reading instruction and comparing it to the traditional method (pg. 2) we stated only one hypothesis -- namely, our prediction that the peer-assisted method of teaching reading would result in higher reading comprehension scores on average than the traditional method.

Even though we stated only one such hypothesis to "go along with" the open-ended problem statement ... whether or not you realized it at the time, there was at least one other "implicit" hypothesis that also "goes with" this particular problem statement!

Or ... to put it another way ... remember our discussion of "retaining" vs. "rejecting" our hypothesis?

Well, if you are led by the study results to "reject" your hypothesis (namely, that the peer-assisted method of teaching reading will result in higher reading comprehension scores) ... AND to 'choose to believe' something else instead ... WHAT is it that you choose to believe?!

This is why hypotheses, whether or not you explicitly write them all out and present them, actually also come in 'sets' for each related research question or problem statement!

They must be:

  1. mutually exclusive (totally separate -- one cannot logically overlap with the other -- e.g., 'either/or') AND

  2. exhaustive ("cover all the bases with regard to reality") -- that is, ONE of them will ALWAYS end up being 'what you ultimately believe about reality' as a result of your study. If you reject one, you will opt to 'retain' or believe ANOTHER one.

    To stick with our example, suppose you're led to reject your initial hypothesis or belief that students who are taught using the peer-assisted reading method will score significantly higher than students taught by the traditional method.

    What will you then 'choose to believe' is true about the entire population?

    This leads us into the concept of "null" hypothesis!

We could have written out the following pair or set of hypotheses corresponding to the problem statement on peer-assisted vs. traditional instruction.

Null Hypothesis
(sometimes written symbolically as H0): Students taught by the peer-assisted method of teaching reading will not score significantly higher on a reading comprehension test than students taught by the traditional method.

Alternative Hypothesis
(sometimes written symbolically as HA): Students taught by the peer-assisted method of teaching reading will score significantly higher on a reading comprehension test than students taught by the traditional method.

Take a break and look at the beginning of the OJ metaphor for research. Hit the forward button at the bottom of this page and have a look at the second page too. We will continue looking at this metaphor throughout the class.

Do you see, first of all, how the above set retains the two key properties of such families?

  • They're mutually exclusive -- they are totally different and can't both be true; AND
  • At the same time, they're exhaustive -- if the first isn't true, then the second automatically must be, and vice versa.

    We might, then, say that the "null" is usually the "rather pessimistic" state of affairs (e.g., "no effect," "no group difference," "no association/relationship," "no impact," & so forth). The null is generally the hypothesis that the researcher will want to reject if he/she anticipates that there "should" be a group difference, relationship, program impact, and so forth.

    (There are some exceptions to this in certain higher-level statistical models & procedures. That is: for those models & procedures, the null is the one 'we want to retain.' But this is more the exception rather than the rule.)

    The alternative hypothesis, then, is one where you are predicting a specific difference, relationship, effect, impact, & so forth.

View more details on null and Alternative hypotheses and examples.

Some Additional Considerations Regarding Hypotheses

  1. Just a 'little subtlety' that may have caught your attention in this lesson packet ... you may have noticed that I referred to "rejecting the null (hypothesis)" if there is compelling evidence to "quit believing it" and go to the alternative. However, you'll also notice that if the opposite is true -- the null seems to be the current state of affairs -- then I didn't say we "accept" the null, but rather that we "retain" it.

    That rather subtle semantic distinction relates to the following point. A "hypothesis" can never be "absolutely proven" per se! There is always the possibility, however slight, that another researcher trying to reproduce your study (we call that "replicating" your study) may come up with new evidence to disconfirm the hypothesis. But if it appears to hold up, rather than assuming it's been proven once & for all,' we 'continue to believe it' -- e.g., we retain it (at least for the time being!).

  2. Also -- it probably makes more sense to formulate hypotheses for certain "families" of problem statements/research questions than others!

    The one family where it might be difficult for you to 'pre-guess' what might happen is the family of "descriptive" questions/statements. You'll recall that these are of the "what is/what are," "identification" focus.

    You simply may not know enough yet about the phenomenon at this point to "pre-guess" your hypotheses! In fact, that's why you're doing the descriptive study in the first place: to find out ('what it is!')!

    So -- it may not be possible, or even desirable, to attempt to formulate and present hypotheses in the case of descriptive research studies.

  3. And that leads us to the final point, which is more of a "personal preference" one ... There simply are no 'hard and fast rules' about whether you need to present BOTH research questions/problem statements AND related hypotheses in your thesis or dissertation! Different committee chairs have different preferences on this issue. The best thing to do would be to discuss this with your thesis/dissertation chair and committee members and ask if they have a preference.

    At a minimum, of course, you need to state the research questions or problem statement. But whether you also need to 'go the extra step' and specify your related hypotheses is a matter of preference, as indicated above.

    Also -- if your thesis/dissertation chair does request that you present your hypotheses along with your problem statement/research question(s) in your study, you may wonder: 1) Do I present just the null? 2) just the alternative? 3) or the complete 'set?' (i.e., the 'pair' of null and corresponding alternative) Here, too, individual preferences vary & there simply are no hard-&-fast rules. Again, I'd advise that you rise the issue directly with your thesis/dissertation chair and committee members if necessary.

    So... to summarize from our three modules thus far:

    Once you have completed this assignment, you should:

    Go on to Assignment 1: Try Your Hand at Writing Hypotheses
    Go back to Understanding Hypotheses

    E-mail M. Dereshiwsky at statcatmd@aol.com
    Call M. Dereshiwsky at (520) 523-1892


    Copyright 1998 Northern Arizona University