Regression help

#1
I ran a regression of event sales (y) and advertising spending (x). I got the following results:

Correlation: .055
R2: .003
N: 438
y = 1.28x + 26492

I believe this verifies my theory that our advertising is not producing much sales. However, events have some contractual advertising. How can I take this into consideration? Do these obligations matter when considering additional marginal sales? Should I even be running a linear regression for this type of analysis?
 

hlsmith

Omega Contributor
#2
I would recommend creating a histogram for both variables then just plotting these data together as well to get a general feel for the relationship.
 

nickx

New Member
#3
Hello,

you might test a moderation effect. You must include in your model the events so that you can determine whether these events significantly influence sales in interaction with the advertising spending.
 

hlsmith

Omega Contributor
#4
Please describe your variables more thoroughly. You have sales at events (y) and "general" advertising (x), what is contractual advertising? Is it specific advertising for the events, not general advertising?
 
#5
I am trying to determine how much of an affect our advertising spending (print, online, tv, etc) has on generating ticket sales to events. All of these events are individual concerts and have their own advertising budget. I plotted a scatter plot of ad spending against ticket revenue and show a minimal relationship. I am questioning the result of the constant I am getting in the equation. It seems unlikely to me that the vast majority of sales would take place with no advertising. Additionally each concert has some advertising requirements so I am trying to take this into consideration.
 

Miner

TS Contributor
#6
Try a plot of ad $s vs. ticket sales by type of advertising. I would expect certain types of advertising to be more/less effective than other forms.
 

hlsmith

Omega Contributor
#7
The type of event may have hetergeneity of effects (slopes) and intercepts. You may want to collapse event into groups and investigate multilevel modeling. Which could control for these in such a model, though the statistics are much much more complex.