41 and median. To know the shape of the distribution of the data we can
use graphs of the distribution.
b. Multicollinearity
Multicollinearity is only to indicate the existence of a linear relationship between variables independent in the regression model.
If the free variables correlated perfectly then it can be called the perfect multicollinearity. To find out whether or not there are
multicollinearity in regression model are as follows: 1 Analyzes the correlation matrix of variables. If there is a
correlation between the free variables are quite high generally above 90 then it is indicated the presence of
multicollinearity. 2 According to Ghozali Priest 2005 : 91 , the value of a
common cutoff is used to indicate the presence of relevant is if the tolerance value is 0.10 or equal to the value of the VIF 10.
This test is done to avoid errors in the estimate the ability of the independent variables as predictors of the dependent
variable. Multicollinearity test done by looking at the value of VIF regression if the value of the independent variable VIF
greater than 10 then it can be inferred that these variables have a very stronglinear relationships with other free variables. With
regard to eliminate Multicollinearity is to eliminate the free variables from the regression equation.
42
c. Heteroscedastisity
Heteroscedastisity indicates that the variation of the variable is not the same for all observations. On Heteroscedastisity
errors that occur are not random, but it shows a systematic relationships in accordance with the size of one or more
variables. Heteroscedastisity aims to test whether the regression model is a variant of the residual Inequality occurs or the observations
of other observations, if the residual variance and one other observation on observations still it is called homocedasticity and if
different is called Heteroscedastisity. The regression models that both homocedastisity and Heteroscedastisity does not occur. To find out or
no Heteroscedastisity there are a number of ways including: 1 See the graph plot between the value prediction variable
ZPRED and residually SRESID. Heteroscedastisity absence detection can be done by looking at the absence of specific
patterns on the chart between SRESID and scatter plot ZPRED where Y is the Y axis that has been predicted and the X axis is
the residual Y – Y predictors actually.
2 Basic analysis, if there is a particular pattern likei points that form a regular pattern wavy, widened, then narrowed. Then,
indication has been Heteroscedastisity. If there is no clear pattern in the points spreads above zero on the Y axis, then does not
happen Heteroscedastisity.
43
3. Multiple Linear Regression Analysis
Multiple linear regression analysis used by the researchers, when researchers intend to predict how the State of the dependent variable
Y, if two or more independent variables as predictors of manipulated factors. So multiple regression analysis would be done when the amount
of a minimum of two independent variables Sugiyono, 2007: 277. the regression Model equations in general form as follows:
Y = a + b1X1 + b2X2 + b3X3 + e
Where: Y
= purchase decisions A
= the number of constants intercept regression b1X
1
= regression coefficient X
1
celebrity endorser the regression coefficient
b2X
2
= X
2
brand image the regression coefficient b3X3
= X 3 consumer perception e
= the standard error
4. T Test Partial
T-test to test the influence of each variable on the variables bound free, then used the following criteria:
Ho = use of celebrity endorser, brand image and consumer perception, it has no effect on purchasing decisions.
Ha = the use of celebrity endorser, brand image and consumer perception affect on purchasing decisions.
44 Decision making criteria t calculate the t table:
1 If t count t table , then Ho accepted. It means the use of celebrity endorser, brand image and consumer perception, it has no effect on
purchasing decisions. 2 If t count t table, then Ho rejaected. It means the use of celebrity
endorser, brand image and consumer perception affect on purchasing decisions.
Decision- making criteria significance with probability α = 0.05:
a. If probability
α 0.05, then Ho accepted. It means the use of celebrity endorser, brand image and consumer perception, it has
no effect on purchasing decisions. b.
If probability α 0.10, then Ho denied. It means the use of celebrity endorser, brand image and consumer affect on
purchasing decisions.
5. F Test Simultan
To test the influence of independent variables toward dependent variable, then use the following criteria:
a. Testing the F with F a countdown table. Ho = Use of celebrity endorser, brand image and consumer
perception, it has no effect on purchasing decisions. Ha = Use of celebrity endorser affects on purchasing decisions.