Limitations for Regressions
Limitations:
Regression analysis is a commonly used tool for companies to make predictions based on certain variables. Even though it is very common there are still limitations that arise when producing the regression, which can skew the results.
The Number of Variables:
The first limitation that we noticed in our regression model is the number of variables that we used. The more companies that you have to compare the greater the chance your model will be significant. We have found that one needs 10-20 times the companies as the variables that are being used (STATSOFT). Our regression was at the low end of this suggestion. We believe that if we had used at least 20-30 more companies our regression line could have been more accurate than what concluded. This was a hard number to come up with because it was found that many of the clothing companies we had were under a parent company. This grouped many of our first choices together under one company. Also, many of the clothing companies are private, therefore limiting the choices we had to find. Since we used the three variable model our limited number of companies made for a better choice than the four or five variable models, as we would have needed more companies to make a more accurate regression line.
Multicollinearity:
Multicollinearity is a limitation problem that is very difficult to avoid. This is known to happen when data located in the x variables are related. When multicollinearity occurs it can cause major problems on the quality and stability of ones final model. (UNESCO.ORG). This was not a huge problem in all of our categories. We did find out that was a severe multicollinearity that occurred between SG&A and the operating income. Because of this we determined that the five variable model would not work for our regression.
Outliers:
When one is comparing a variety of companies it is discovered that there is a difference in size from one company to another, even
Regression analysis is a commonly used tool for companies to make predictions based on certain variables. Even though it is very common there are still limitations that arise when producing the regression, which can skew the results.
The Number of Variables:
The first limitation that we noticed in our regression model is the number of variables that we used. The more companies that you have to compare the greater the chance your model will be significant. We have found that one needs 10-20 times the companies as the variables that are being used (STATSOFT). Our regression was at the low end of this suggestion. We believe that if we had used at least 20-30 more companies our regression line could have been more accurate than what concluded. This was a hard number to come up with because it was found that many of the clothing companies we had were under a parent company. This grouped many of our first choices together under one company. Also, many of the clothing companies are private, therefore limiting the choices we had to find. Since we used the three variable model our limited number of companies made for a better choice than the four or five variable models, as we would have needed more companies to make a more accurate regression line.
Multicollinearity:
Multicollinearity is a limitation problem that is very difficult to avoid. This is known to happen when data located in the x variables are related. When multicollinearity occurs it can cause major problems on the quality and stability of ones final model. (UNESCO.ORG). This was not a huge problem in all of our categories. We did find out that was a severe multicollinearity that occurred between SG&A and the operating income. Because of this we determined that the five variable model would not work for our regression.
Outliers:
When one is comparing a variety of companies it is discovered that there is a difference in size from one company to another, even
References: http://www.unesco.org/webworld/idams/advguide/Chapt5_2.htm http://www.statsoft.com/textbook/multiple-regression/#alimitations http://www.advfn.com