Page 105 -

4.7 Identifying relationships 91

4.7.2 REGRESSION
Unlike correlation analysis, which allows the study of only two variables, regression
analysis allows you to investigate the relationship among one dependent variable
and a number of independent variables. In HCI-related studies, regression analysis
is used for two main purposes: model construction and prediction. In cases of model
construction, we are interested in identifying the quantitative relationship between
one dependent variable and a number of independent variables. That is, we want to
find a mathematical equation based on the independent variables that best explains
the variances in the dependent variable. In cases of prediction, we are interested in
using a number of known factors, also called “predictor variables,” to predict the
value of the dependent variable, also called the “criterion variable” (Share, 1984).
The two objectives are closely related. You need to build a robust model in order to
predict the values of the criterion factor in which you are interested.
Depending on the specific research objective, you need to choose different re-
gression procedures to construct the model. If the objective of the study is to find
the relationship between the dependent variable and the independent variables as a
group, you can enter all the independent variables simultaneously. This is the most
commonly adopted regression procedure (Darlington, 1968). Using this approach,
you will find out the percentage of variances in the dependent variable that can be ex-
plained by the independent variables as a group. This percentage is presented in the
2
2
form of R . If the procedure returns a significant R , it suggests that the independent
variables as a group have significant impact on the dependent variable. This proce-
dure is useful but is insufficient if you are interested in the impact of each individual
independent variable.
If you want to create a model that explains the relationship between the depen-
dent variable and each individual independent variable, the hierarchical regression
procedure is appropriate. Using this procedure, you will enter the independent vari-
ables one at a time into the regression equation. The order of the entry of the indepen-
dent variables is determined by the predefined theoretical model. The independent
variables that are entered into the equation first usually fall into two categories. One
category includes variables that are considered to be important according to previ-
ous literature or observation; in this case, you want to evaluate the overall impact of
this variable on the dependent variable. The second category includes the variables
that are of no interest to you but have significant impact on the dependent variable
(also called covariates); in this case, you want to exclude the variable's impact on the
dependent variable before you study the variables that you are interested in. In other
words, entering the covariates first allows you to remove the variances in the depen-
dent variable that can be explained by the covariates, making it easier to identify
significant relationships for the variables in which you are interested.
Suppose you conduct a user study that investigates target selection tasks using a
standard mouse. One important dependent variable of interest is the task completion
time and you want to know what factors have an impact on task completion time.
There are a number of potential factors such as target size, distance, computer experi-
ence, age, etc. In order to find the relationships among the factors, you can conduct