4

I have questions about a study I was reading. In this study about children and money, the relationship between the age of the children and their weekly pocket money was assessed. The researchers studied the relationship between age and pocket money on a scatter plot and added a linear regression model.

  • Why did they use a scatter plot?
  • Could they have used something else?
Jeromy Anglim
  • 42,044
  • 23
  • 146
  • 250
Mimi
  • 299
  • 2
  • 5
  • 8

3 Answers3

3

For this I would expect a wide variation, as well as an increasing trend with age. If there is a lot of data then you could use boxplots (or some of the modern variants such as violin plots) segmented by age.

Henry
  • 30,848
  • 1
  • 63
  • 107
2

When looking at the relationship between two continuous variables, a scatterplot is very often a very good idea. In this article, the authors DID do something else: They looked at a linear regression model. They could do things besides that, although it's hard to tell what would be good, without more information. After running this model, they should check assumptions: Graphics are often very valuable here.

If looking at more than two variables (which is usually a good idea) then there are a number of alternative plots.

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
  • 1
    I am wondering that which plots can be used as alternative to scatter-plot? – love-stats Oct 05 '11 at 23:18
  • 1
    Well, there are trellis plots, which are really a variation of scatter plots, sort of. Scatterplot matrices. For hypervariate data, a lot of things have been proposed; none of them nearly perfect, in my view. – Peter Flom Oct 05 '11 at 23:52
  • 1
    Peter, I wonder why do you think a scatter plot is very often a very good idea when looking at the relationship between two continuous variables? – Mimi Oct 06 '11 at 17:08
  • 1
    @mimi They can show all sorts of relationships: Linear, nonlinear, nonmonotonic... They can show outliers. They show ALL the data. Additions such as loess or other smoothed lines help, and if there are a lot of points, other methods may be better. I wrote a paper about scatterplots for NESUG, it is here http://www.nesug.org/Proceedings/nesug11/po/po06.pdf (this uses SAS, but could easily be duplicated in R or another package). – Peter Flom Oct 07 '11 at 10:37
2

As an example of @Henry's mentioned "some variants" there are dot-plots which often look good. It is actually a sliced scatterplot showing conditional distributions of money at various age levels.

enter image description here

ttnphns
  • 51,648
  • 40
  • 253
  • 462