Scatterplot Question




I have designed two Likert scales to assess the purchase likelihood of a brand with a certain attribute and the second one to assess the attitude towards this attribute (both are 7-point scales). I based my questionnaire on the theory of planned behaviour which says that attitude is one of the factors that influences the purchase intention. Now, I want to see if there is a relationship between the purchase intention and the attitude so I'm going to do a Spearman Rank Order Correlation. I first generated a scatterplot but I'm not sure if it looks correct. To me, it looks linear with a positive slope and a strong relationship. I've checked common patterns for scatterplot but none of them looks like the one I have.

This is why I would like to know if anyone of you would be able to help to read my scatterplot. I've done this for 9 other attributes and the graphs look more or less similar.

Thank you very much for you help.


PS: I've attached the graph to the thread.
You really can't tell based on the way the plot is formatted. Since there are no data points in the upper left-hand corner or lower right-hand corner, it might suggest a positive slope or at least not a negative one, but you can't be sure. Most stats programs have features you can add to the plot that would be helpful, such as a fit line, basing the size of the points on the number of respondents, and indicating what the correlation (r) (or r squared) is on the chart. If there's a strong relationship, the fit line will go diagonally from the lower left-hand corner to upper right-hand corner (or vice-versa, if it's a negative relationship), and the largest data points will be close to that line.


Less is more. Stay pure. Stay poor.
I have myself produced similiar graphs (very very ambiguous). The issue with the ordinal data and this graph is multiple observations at the same point cannot be seen or distinguished (since they are on top of each other). Some graphs may allow the dots to change size based on the number of observations at that location. You are better served looking at the output, not graph in this situation. What is the correlation (r) and respective p-value?


Ambassador to the humans
It would be much better to have the size of the dot correspond to how many points fall at that location. Although I also like adding a small amount of "jitter" so that the points aren't directly overlapped.


Ambassador to the humans
I don't know how to do it with an option - but I don't use SAS for plotting. You could create a modified dataset though and do the plotting based on that.