Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Percent or percentage point?

    Hi,
    I have a regression where I have return on assets as the dependent variable, and a set of dummy variables, and accounting variables as the independent variables. I run this regression in STATA and get a coefficient of 8.5 for the dummy variable "wealth". (Mening the person owning the company is wealthy). My question is, does this mean that wealth effects ROA by 8.5% or by 8.5 percentage points?

    Thank you!

  • #2
    Percentage points.

    Remember, if you do linear regression, you essentially have ROA = b*wealthDummy + e
    Imagine ROA of a poor person is 3.5%, then the ROA of a wealthy person is 3.5 + b.
    Your estimate for b is 8.5, which implies a ROA of 3.5 + 8.5 = 12

    Comment


    • #3
      Thank you Jesse! Does this even hold when it is a panel data?
      Sorry for stupid question, I'm new with stata

      Comment


      • #4
        Yes! And no problem (you'd be surprised how many published academic papers write percentage when it should be percentage points...)

        Comment


        • #5
          https://xkcd.com/985/

          Click image for larger version

Name:	percentage_points.png
Views:	2
Size:	32.4 KB
ID:	1454455
          David Radwin
          Senior Researcher, California Competes
          californiacompetes.org
          Pronouns: He/Him

          Comment


          • #6
            If the units of ROA are percent, wouldn't it be acceptable to assume that you can say, "The intercept is 3%, and owner wealth is associated with an additional 8.5%" and expect the listener to understand that we're talking about a total ROA of about 12%? To distinguish a proportional increase of (8.5 + 3) / 3, wouldn't you be expected to say "a fourfold increase" or "an increase of 400% of baseline", the qualification being necessary in the context of discussion of a linear model where the assumption is additivity?

            I work with industrial product-development studies all of the time, where it is common to express results in terms of rates and risk differences. I don't recall ever seeing the term "percentage point" in any protocol, statistical analysis plan or study report. No one I've spoken with has appeared to misunderstand what a "10% margin of noninferiority" means.

            Is it an econometrics thing? Where it's often the log of this versus the log of that in a linear regression model, and so discussion of regression coefficients can get muddled unless you're explicit?
            Last edited by Joseph Coveney; 20 Jul 2018, 20:04.

            Comment


            • #7
              The increase, itself, would be threefold, and not four,but that would underscore the need to stick with the metrics of the model when describing the results.

              Comment


              • #8
                I suppose it's field-dependent indeed. If your field almost exclusively deals with absolute changes in percentage-valued variables, then using percentage points instead of just % is a waste of energy. Alternatively, in a field like finance you often have stocks going up 0.5% (percent), but also interest rates going up by 0.5% (percentage points). In the end, if you are familiar with the context you can almost always deduce from the context which of the two is relevant. But for example, I have no idea whether a 10% margin of noninferiority would refer to 10pp or 10%, but that's mainly because I haven't the faintest idea what it is. In academia this is quite a common situation - the authors might be talking about concepts you've only learned about a few pages back and then you don't necessarily know whether that 8.67% increase is really a percent or a percentage point.

                Comment


                • #9
                  Thank you for sharing your perspective.

                  Comment

                  Working...
                  X