Level of detail in hypotheses

Discussion in 'General Science & Technology' started by Michiel, Jul 28, 2011.

Thread Status:
Not open for further replies.
  1. Michiel Registered Member

    Messages:
    2
    Hi all, I'm new here, hopefully posting this in the right section.

    I'm currently writing a paper incorporating results from survey research. I have tested various hypotheses, but I am unsure what level of detail I should employ in my writing. If I would include every hypothesis I've tested, my paper would easily have 40 hypotheses or more. However, if I would simplify the hypotheses (group them) I fear they are not provable or falsifiable anymore...

    For example, I've tested the likeliness of people using one software system also using another. Thus, H0 reads "People using a certain software system are more likely to use another one as well than people who don't use any." Because I have several systems, however, I have to conduct several tests on my dataset which yields multiple results; if, then, only some of those results are significant, should I 'partially accept' the hypothesis or reject it altogether...? The alternative would be, of course, to include a specific hypothesis for every single system, but this would enormously expand the number of hypotheses and make my paper unreadable.

    How should I cope with this? Thanks for the insight.

    Please Register or Log in to view the hidden image!

     
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. MRC_Hans Skeptic Registered Senior Member

    Messages:
    835
    When you test a hypothesis, you look at which prediction it makes, i.e. (seemingly in your vein): "Users of software A tend to also use software B" will mean that for a given number of A users the incidence of B will be higher than among a similar number of non-A users.

    Of course, to make sense, it requires that there are no practical interdependencies. For instance, you will not need to test the hypothesis that MSIE is predominantly used by Windows users.

    When you have a valid and falsifiable prediction, you proceed to design a test around it. In this case you compare a group of A users with a group of non-A users with the respect to their use of B. If a significantly higher number of A users use B, then the hypothesis is supported, otherwise it fails.

    However, it seems that you have instead gathered a large amount of data, of several kinds. This method is a form of basic research, and you can now approach it in two different ways:

    1) For each hypothesis, like above, design a proper test, extract relevant data from your data set, and test it. You will write a report on each hypothesis. Be careful to use all relevant data.

    2) Perform a statistical analysis of the whole data set, to map interdependencies using a statistical multivariate tool. However, if you have a lot of variables, this becomes quite hairy.

    Hans
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. Michael 歌舞伎 Valued Senior Member

    Messages:
    20,285
    I always suggest to beginning students to write their hypothesis like this:

    If (dependent variable) is related to (independent variable) then (prediction).
     
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. Fraggle Rocker Staff Member

    Messages:
    24,690
    I would recommend rewriting that, since "related to" can refer to correlation, and (presumably) what you're looking for is causation.

    The fallacy of correlation is, perhaps, not the most common fallacy, but it is arguably the most insidious since many experiments are not designed to distinguish correlation from causation.
     
  8. Michiel Registered Member

    Messages:
    2
    Thanks for your answers, guys.

    As much as I'd like to, I cannot distribute the different findings over multiple papers; on their own, there wouldn't be enough context or body to justify publication...

    Please Register or Log in to view the hidden image!



    However, I've now worked on generalizing the hypotheses by phrasing them singular and removing the specific systems' names. That leads to a reasonable number of hypotheses; for all possibilities I now only have to explicitly state whether or not that hypothesis holds.

    One of my hypotheses reads, e.g.: 'Usage of a decision-supportive system (DSS) positively influences usage of a medical system (MS).'
    I then specify that it holds for DSS A & MS A, DSS B & MS A, but not for DSS A & MS B.
     
  9. Michael 歌舞伎 Valued Senior Member

    Messages:
    20,285
    In this case, that being a beginning student, I think "is related" reminds students that they are investigating relationships, not just cause and effect events.

    For example: If potato rigidity is related to turgor pressure, then increasing (or decreasing) the water content will increase (or decrease) the rigidity of cut potato strips.

    The causative relationship is identified by the experimental design which is stated in the prediction.

    Lets see what happens if it's purely correlation.

    If murders are related to ice cream sales, then increasing ice cream sales will increase murder rate.

    Well, we know that increasing ice cream sales increases murder rate BUT once we set control variables (sunny days) Ha! The correlation disappears

    Please Register or Log in to view the hidden image!

    and we can see it's actually not a causative relationship IN the experimental design.
     
  10. Michael 歌舞伎 Valued Senior Member

    Messages:
    20,285
    Also, wouldn't you potentially miss a relevant finding if you rejected the alternate hypothesis because it wasn't causative? What if there's a two way interaction or something else interesting going on?
     
    Last edited: Aug 1, 2011
  11. Pinwheel Banned Banned

    Messages:
    2,424
    How confusing. Michiel ? Michael ?
     
  12. Cifo Day destroys the night, Registered Senior Member

    Messages:
    685
    IMO, the statement of a hypothesis, theory, etc should be kept as brief and precise as possible; however, the description can also include limitations and other exceptions. What separates hypotheses from other explanations is the lack of purpose and rigor in collecting the data known at that time.

    For example, the equation PV=nRT could have been a hypothesis at some point in history, even though it is very precise and is now known as a scientific "law". To form this hypothesis, someone may have combined observations from other studies, such as:
    • the relationship now called Avogadro's Law (volume, V, is proportional to moles, n)
    • the relationship now called Boyle's Law (pressure, P, is inversely proportional to volume, V)
    • the relationship now called Charles's Law (volume, V, is proportional to temperature, T)
    However, no one had yet to test the hypothesis of PV=nRT with experiments designed to purposely monitor all these variables under rigorously controlled conditions. IMO.
     
  13. Michael 歌舞伎 Valued Senior Member

    Messages:
    20,285
    Mee kel

    Maa ee kel

    How's that?

    Please Register or Log in to view the hidden image!

     
  14. Michael 歌舞伎 Valued Senior Member

    Messages:
    20,285
    I suppose it also depends on the field you're doing your research into?
     
  15. MRC_Hans Skeptic Registered Senior Member

    Messages:
    835
    Ni shi Zongguoren ma?

    Hans
     
Thread Status:
Not open for further replies.

Share This Page