0% found this document useful (0 votes)
275 views29 pages

Regression Step-by-Step JASP Guide

The document describes a study that examined the relationship between employees' cultural intelligence and innovative behaviors at work. It discusses how the researchers measured cultural intelligence and innovative behaviors using Likert scale questions. The document then explains that a linear regression will be conducted to determine if cultural intelligence predicts innovative behaviors. It provides the hypotheses and equation for the linear regression. Finally, it describes testing the assumptions of linear regression by examining the normality, outliers, linearity, and heteroscedasticity of the variables.

Uploaded by

Ivana Koleva
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
275 views29 pages

Regression Step-by-Step JASP Guide

The document describes a study that examined the relationship between employees' cultural intelligence and innovative behaviors at work. It discusses how the researchers measured cultural intelligence and innovative behaviors using Likert scale questions. The document then explains that a linear regression will be conducted to determine if cultural intelligence predicts innovative behaviors. It provides the hypotheses and equation for the linear regression. Finally, it describes testing the assumptions of linear regression by examining the normality, outliers, linearity, and heteroscedasticity of the variables.

Uploaded by

Ivana Koleva
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 29

Running and Interpreting Linear Regression in JASP

We are going to use a portion of the data that was used in the study by Li, Wu, and Xiong
(2021). This is the same study we used when we conducted a Pearson Correlation in the previous
guide and found a strong, positive relationship between employee levels of cultural intelligence
(e.g., ability to adapt to cross-cultural environments) and their innovative behaviors in the
workplace. For this guide, we will be building upon that analysis by first conducting a simple
linear regression to determine if Cultural Intelligence predicts innovative behaviors in the
workplace. As we learned in the previous guide, Sustainable Innovation Behavior refers to how
well an individual creates and enacts new ideas.

The researchers measured Sustainable Innovation Behavior (SIB) by asking participants to


answer six questions rated on a 7-point Likert scale. Higher scores on this scale indicate higher
levels of Sustainable Innovative Behavior. A sample item includes, “I often come up with
creative ideas.”

The researchers measured Cultural Intelligence (CQ) by asking participants to answer 12


questions rated on a 7-point Likert scale. Higher scores on this scale indicate higher levels of
Cultural Intelligence. A sample item includes, “I adjust my cultural knowledge as I interact with
people from a culture that in unfamiliar to me.”

Now we are ready to try out an analysis of our own. Open up the “Li et al 2021” file. You have
both a .csv file you can upload and a JASP file.
Hypotheses
We know there is a relationship between Cultural Intelligence and Sustainable Innovation
Behaviors after completing a Pearson correlation; however, now we are going to build on those
results to determine if Cultural Intelligence predicts Sustainable Innovation Behaviors using a
linear regression model.

The null hypothesis is:


● Conceptual H0: Cultural Intelligence will not predict Sustainable Innovation Behavior.
● Mathematical H0: b = 0

The alternative hypothesis is:


● Conceptual H1: Cultural Intelligence will predict Sustainable Innovation Behavior.
● Mathematical H1: b ≠ 0
Linear Regression Equation
Y′ = bX + a
Sustainable Innovative Behavior′ = (b x Cultural Intelligence) + a
Here is a breakdown of the components of the regression question:

1|Page
• Y′ is the predicted value of your outcome or dependent variable (Y)
• b is the slope of the regression line (estimated by regression)
o May also be written using this symbol instead: ꞵ1
• X is the value of your predictor variable or independent variable
• a is the Y-intercept of the regression line (estimated by regression)
o May also be written using this symbol instead: ꞵ0
Assumption Testing
Let’s walk through testing all six of our assumptions.
Assumption One: Are our variables continuous?
Yes. Our predictor, Cultural Intelligence is a continuous variable measured using a 12-item
Likert scale questionnaire. Our dependent variable is also a continuous variable measured using a
6-item Likert scale questionnaire.
Assumption Two: Are the variables normally distributed?
To continue our assumption testing, click Descriptives. When the “Descriptive Statistics”
window pops up, we will need to move the following study variables to the “Variables” box on
the right: Cultural Intelligence and Sustainable Innovation Behavior.

In the Statistics drop down menu, check the boxes next to Skewness and Kurtosis under
Distribution.

2|Page
Under Basic Plots select Distribution Plots. In the Customizable Plots drop down menu you want
to select several options for the next set of assumptions.
1. Under Boxplots select Label outliers
2. Under Scatterplots
a. Select None under Graph above scatter plot
b. Select None under Graph right of scatter plot
c. Select Add regression line
i. Under Add regression line select Linear and Show confidence interval
95.0%

3|Page
Now let’s look at our results. We want histograms that look approximately symmetrical as well
as skewness and kurtosis statistics between -2 and +2 for each of our study variables.

4|Page
Looking at our results in the preceding images, we see that our skewness and kurtosis values are
all within the acceptable range of -2 to +2. Our histograms also confirm that our variables are
relatively normally distributed, with approximately symmetrical distributions; however, the way
the histograms for each of the variables have a slight tail to the left indicates we may have
outliers present in the bottom quartiles of these distributions. To confirm, let’s move to our next
assumption.

To report these results in APA format, we can write:


The scores on both of our study variables were normally distributed, with skewness and
kurtosis statistics within the acceptable range of -2 and +2.
Assumption Three: Are there any outliers?
To determine if there are any outliers within any of our study variables, we will look at the
boxplot output in the following image. As suspected from the preceding histograms, there are
outliers present in the bottom quartiles of both variables. There are two outliers in the
Sustainable Innovative Behaviors variable and two outliers present in the Cultural Intelligence
variable. As we have discussed before, researchers have to decide whether they are going to keep
or delete outliers from their analyses. We will complete our analyses with our outliers for now,
but it is good practice to rerun analyses with outliers deleted afterwards to determine if the
results of your analyses would be different.

To report this using APA format, we would write:

5|Page
There were outliers identified in the bottom quartile of the cultural intelligence and
sustainable innovative behaviors distributions, as assessed by an inspection of boxplots.
There were two outliers in the sustainable innovative behaviors distribution and two
outliers present in the cultural intelligence distribution.
Assumption Four: Are the relationships of interest linear?
Now we need to look at the scatterplot we requested in our output. Looking at the following
scatterplot, our predictor variable (Cultural Intelligence) seems to have a linear relationship with
our outcome variable, Sustainable Innovation Behavior.

To report this using APA format, we would write:


There is a linear relationship between cultural intelligence and sustainable innovative
behavior, as visualized using a scatterplot.

Assumption Five: Are there concerns with heteroscedasticity?


Looking back at the preceding scatterplot, there is not a concern for any cone-shaped
distributions. Though there is certainly variability along our best-fit line, there is not one area of
the line that has drastically different levels of variability than another. In other words, our error
variances or residuals seem to be relatively similar across all values of our predictors.

To report this using APA format, we would write:


There are no concerns regarding heteroscedasticity, after a visual inspection of a
scatterplot.

6|Page
Primary Analyses
Now we are ready to run our linear regression. We can find the regression analysis with the
regression tab, which we also used for correlations. We will select the option for linear
regression.

In the main regression menu, we will move over Sustainable Innovation Behavior as our
dependent variable and Cultural Intelligence as our covariate (or predictor variable).

There is nothing we need to adjust in the “model” drop down, so we will move on to “statistics”.
In addition to the default settings, we will click:
7|Page
● “Confidence intervals” under regression coefficients to get a range of likely slope values.
● “Statistics” under residuals to look at the typical amount of error in our model.
● “Descriptives” in the right column of options, just to look at typical values in our dataset.
● “Durbin-Watson” to check assumption six, that we have independence of errors.

We will also skip over the model specification dropdown and move onto plots. In the plot
dropdown, we will click on Residuals vs. histogram to request residuals plotted as a histogram.

8|Page
Now we’re ready to look at our output! First let’s begin by checking our remaining assumption.
Assumption Five: Do we have independence of errors?
One important consideration in regression analysis is whether the residuals (the differences
between the actual values and predicted values) are correlated with each other or not. The
Durbin-Watson statistic is a measure of this correlation. When the Durbin-Watson statistic is
between 1.5 and 2.5, it suggests that there is not a significant correlation between the residuals or
errors in our model. This is important because if there is a high degree of correlation between
residuals, it can suggest that the model is not capturing all the relevant information about the
relationship between the variables. In other words, the model may not be a good fit for the data.
A Durbin-Watson statistic outside of the 1.5-2.5 range may indicate that there is a problem with
the model, and further investigation may be needed to determine the cause of the correlation
between residuals.
The Durbin-Watson statistic is a part of the Model Summary results. If this value is between 1.5-
2.5, we are not concerned there is a significant amount of correlation between the residuals.

Looking at the preceding figure, we can determine that our Durbin-Watson statistic is within the
acceptable range of 1.5-2.5. To report this in APA format:

The model has acceptable independence of errors, Durbin-Watson = 1.92.

Now that we have finished checking our assumptions, we can move onto the main results
displayed in the first three tables. Notice the tables have a line for H0 and H1. This estimates the
variance in the outcome with no predictors (H0) and then adds a predictor in line with the
alternative hypothesis (H1). We will be focusing on H1 output.

The model summary table tells us about the variance explained by our predictor variable,
Cultural Intelligence. Specifically, we want to look at the R and R2 values. The R value is a
correlation coefficient that tells us that Cultural Intelligence has a strong, positive correlation
with Sustainable Innovation Behavior (r = .669). Looking at our effect size, r2, we can also say
that Cultural Intelligence explains 44.8% of the variance in Sustainable Innovation Behavior.

9|Page
Another way of saying that is that 44.8% of the variance in Sustainable Innovation Behavior can
be accounted for by Cultural Intelligence.

The next part of the results we will be looking at is the ANOVA table. The ANOVA table tells
us if our best-fit line explained a big enough portion of variance in the relationship between our
variables to be statistically significant. For the variables we have in our model, it is a significant
F value because our probability value is smaller than .05 (F = 270.69, p < .001).

To report this using APA format, we would write:


Cultural Intelligence significantly predicts Sustainable Innovation Behavior, F (1, 334) =
270.69, p < .001, r2 = .45. Cultural Intelligence explains 44.8% of the variance in
Sustainable Innovation Behavior.

Typically, we care more about the individual regression coefficients (slopes) than the overall
model significance. To find that information, we move to the last table, labeled coefficients. In
the coefficients table, the first column, labeled unstandardized, provides the intercept in the first
row of the H1 output and the slope for Cultural Intelligence in the second row of output. We see
that our intercept is 9.31, meaning that if a person scored 0 on CQ, the average Sustainable
Innovation Behavior score of that person would be around 9.31. Our slope is .37, meaning that
for every one-unit increase in Cultural Intelligence, people are expected to score .37 more on

10 | P a g e
Sustainable Innovation Behavior. Putting this together, we could write out our regression
equation:

Sustainable Innovation Behavior′ = .37 (Cultural Intelligence score) + 9.31

We are also given an estimate of error around our slope and intercept in the second column. This
error term is used to compute the t value. For example, if you divide the slope by the standard
error (-.369 / .022) you would get our t value of 16.45. Our t value of 16.45 has an associated p
value of < .001. Because this is less than our alpha level (.05), this tells us that the slope is
significantly different from zero. This means there is a significant relationship between Cultural
Intelligence and Sustainable Innovation Behavior, b = .37, SE = .02, t (334) = 16.45, p < .001.
[Note: the df for t is the same as the degree of freedom reported in the residual row on the
omnibus ANOVA table.]

Some additional pieces of output to notice are the standardized estimate and the confidence
intervals. The standardized column gives us beta, which is the standardized slope. These values
range from 0 to 1, just like a correlation. You may have noticed the standardized slope is the
same as our R value in the Model Summary. These values will be the same if you only have one
predictor variable. The confidence interval tells us a range of likely slope values within the
population. If we sampled 100 times, 95 of those times, we would expect a slope estimate
between .33 and .41.

The next couple of tables can provide some helpful context. Our descriptive statistics table tells
us about typical levels of Cultural Intelligence and Sustainable Innovation Behavior in our
sample. For instance, the average Sustainable Innovation Behavior in the whole sample was
29.9.

11 | P a g e
The residuals table tells us about the typical levels of discrepancy between the expected value of
Sustainable Innovation Behavior (Y) given the Cultural Intelligence score, and the actual value
reported in our data. We can see that our average residual is very small and there’s a range of
prediction error, with the maximum underestimate (shown as the minimum) as -14.12 and our
biggest overestimate, being off by about 13.51. The average standardized residual value tells us
that we were often off by a very small amount, technically .000046 standard deviations. This
information can be informative, but a plot is more helpful for viewing our residuals.

Our plot of the standardized residuals (differences between predicted Y and actual Y score) show
a relatively normal distribution. This visual, helps us to be confident that are residuals are
normally distributed, which is one of the assumptions we need to be true as we examine our
prediction error.

12 | P a g e
Putting it all Together
What do you need to report in your results?

Value (Notation)
slope (b)
standard error of slope (SE)
t value (t)
Degrees of Freedom (df)
Probability (p)
Effect Size (r2 or R2)

Assumptions
Skewness and Kurtosis
Histogram or Boxplot
Scatterplot
VIF/Tolerance
Plot of residuals

It’s time to summarize the results of our regression analysis in APA format.

Prior to conducting a simple linear regression, we tested several assumptions. We examined our
predictor (Cultural Intelligence) and outcome (Sustainable Innovation Behavior) variables for
normality and outliers. Skewness and kurtosis values were all in the normal range and
visualizing the variables with boxplots and histograms revealed four outliers that were retained
for analysis. Using a scatterplot, we also confirmed a linear relationship between the variables
with no evidence of heteroscedasticity. The residual histogram showed the residuals were
approximately normally distributed.

13 | P a g e
For our primary analyses, we examined how Cultural Intelligence predicted Sustainable
Innovation Behavior. The overall model was significant, F (1, 334) = 270.69, p < .001, r2 = .45.
There was a significant relationship between Cultural Intelligence and Sustainable Innovation
Behavior, b = .37, SE = .02, t (336) = 16.45, p < .001. Our slope is .37, meaning that for every
one-unit increase in Cultural Intelligence, people are expected to score .37 higher on Sustainable
Innovation Behavior. Cultural Intelligence explained 44.8% of the variance in Sustainable
Innovation Behavior.

Running and Interpreting Multiple Regression in JASP


Now that we have the basics down, let’s add additional predictors used in Li and colleagues
(2021) original model to conduct a multiple regression. To do this, we will be conducting a
multiple regression to examine how and if Cultural Intelligence, Organizational Culture
Differences, and Knowledge Sharing have an impact on our dependent variable, Sustainable
Innovation Behavior. Why are we adding additional variables? Most of the time our outcomes
are affected by more than just one thing, so we can be more accurate in predicting if we include
more than one factor that may be relevant. Organizational Culture Differences refers to a
company’s corporate culture and measures three aspects: values, systems, and management
behaviors. Knowledge Sharing refers to an individual’s willingness to share their cultural
knowledge with others in the organization.

The researchers measured Organizational Culture Behavior (OCB) by asking participants to


answer 5 questions rated on a 7-point Likert scale. Higher scores on this scale indicate higher
levels of Organizational Culture Behavior. A sample item includes, “Organizations uses
consensus seeking rather than authoritarian decision making.”

The researchers measured Knowledge Sharing (KS) by asking participants to answer 4 questions
rated on a 7-point Likert scale. Higher scores on this scale indicate higher levels of Knowledge
Sharinh. A sample item includes, “The more knowledgeable members will provide the other
members with knowledge or skills that are difficult to acquire for free.”
Hypotheses
Our hypothesis concerns whether we can predict Sustainable Innovation Behavior using three
separate predictor variables: Cultural Intelligence, Knowledge Sharing, and Organizational
Culture Differences.

The null hypothesis is:


● Conceptual H0: Cultural Intelligence, Knowledge Sharing, and Organizational Culture
Differences will not predict Sustainable Innovation Behavior.
● Mathematical H0: b = 0

The alternative hypothesis is:

14 | P a g e
● Conceptual H1: Cultural Intelligence, Knowledge Sharing, and Organizational Culture
Differences will predict Sustainable Innovation Behavior.
● Mathematical H1: b ≠ 0
Assumption Testing
Let’s walk through testing all six of our assumptions.
Assumption One: Are our variables continuous?
Yes. Our predictors, Cultural Intelligence, Knowledge Sharing, and Organization Culture are all
continuous variables measured using multiple item Likert scale questionnaires. Our outcome of
Sustainable Innovation Behavior is also continuous.
Assumption Two: Are the variables normally distributed?
To continue our assumption testing, click Descriptives. When the “Descriptive Statistics”
window pops up, we will need to move the following study variables to the “Variables” box on
the right: Cultural Intelligence, Knowledge Sharing, Organizational Culture Differences, and
Sustainable Innovation Behavior.

In the Statistics drop down menu, check the boxes next to Skewness and Kurtosis under
Distribution.

15 | P a g e
Under Basic Plots select Distribution Plots. In the Plots drop down menu you want to select
several options for the next set of assumptions.
1. Under Boxplots select Label outliers
2. Under Scatterplots
a. Select None under Graph above scatter plot
b. Select None under Graph right of scatter plot
c. Select Add regression line
i. Under Add regression line select Linear and Show confidence interval
95.0%

16 | P a g e
Now let’s look at our results. We want histograms that look approximately symmetrical as well
as skewness and kurtosis statistics between -2 and +2 for each of our study variables.

17 | P a g e
Looking at our results in the preceding images, we see that our skewness and kurtosis values are
all within the acceptable range of -2 to +2. Our histograms also confirm that our variables are
relatively normally distributed, with approximately symmetrical distributions; however, as we
pointed out with the simple regression, the way the histograms for each of the variables have a
slight tail to the left indicates we may have outliers present in the bottom quartiles of these
distributions. To confirm, let’s move to our next assumption.

To report these results in APA format, we can write:


The scores on all four of our study variables were normally distributed, with skewness
and kurtosis statistics within the acceptable range of -2 and +2.
Assumption Three: Are there any outliers?
To determine if there are any outliers within any of our study variables, we will look at the
boxplot output in the following image. As suspected from the preceding histograms, there are
outliers present in the bottom quartiles of each of the four variables. There are two outliers in the
Sustainable Innovative Behaviors variable, two outliers present in the Cultural Intelligence
variable, one outlier in the Knowledge Sharing variable, and several outliers present in the
Organizational Cultural Differences variable.

18 | P a g e
To report this using APA format, we would write:
There were outliers identified across all four study variables in the bottom quartile of the
distributions. There were two outliers in the Sustainable Innovation Behavior distribution,
two outliers present in the Cultural Intelligence distribution, one outlier in the Knowledge
Sharing distribution, and several outliers present in the Organizational Culture
Differences distribution.
Assumption Four: Are the relationships of interest linear?
Now we need to look at the scatterplots we requested in our output. Looking at our scatterplots,
all of our predictors (Cultural Intelligence, Knowledge Sharing, and Organizational Culture
Differences) seem to have a linear relationship with our outcome variable, Sustainable
Innovation Behavior.

19 | P a g e
Assumption Five: Are there concerns with heteroscedasticity?
Again, looking at our scatterplots, there is not a big concern for any cone-shaped distributions.
Though there is certainly variability along our best-fit line, there is not one area of the line that
has drastically different levels of variability than another. In other words, our error variances or
residuals seem to be relatively similar across all values of our predictors.

Related, we will also take a closer look at our residuals to see if they are normally distributed
when we get to our actual regression analysis and output.
Assumption Six: (For the multiple regression only) Are our predictor variables highly
correlated?
If our predictor variables are highly correlated with each other, this can create problems in
interpreting the results of the analysis. Collinearity is the term used to describe this issue of high
correlation between predictor variables. One way to check for collinearity is by looking at the
correlations between the variables. Another method is by checking the Variance Inflation Factor
(VIF) and Tolerance statistics. Let’s begin by using what we learned in the correlation exercise,
let’s click the regression menu, select correlation, and then select our four variables. We are
interested in looking at the Pearson’s r values for our study variables, so we won’t select any
fancy options this time.
In our output table, we see that all three predictors are related to our outcome. We will consider
those relationships in more detail with our actual regression analysis. For now, we are concerned
with whether our three predictors are too highly correlated. In this case, the Cultural Intelligence
and Knowledge Sharing are correlated at .63, meaning that the more cultural intelligence, the
more knowledge sharing there is. This correlation is strong, so it is high enough that we may be
concerned about multicollinearity in our data. Cultural Intelligence and Organization Culture
Difference are moderately correlated at .52 and Knowledge Sharing and Organization Culture

20 | P a g e
Difference are also highly correlated at .62. Keep in mind, we will also look at the VIF and
tolerance values in the regression output as additional information about this assumption.

Primary Analyses
Now we are ready to run our multiple regression model. We will go back up to the regression tab
at the top of the JASP menu bar and select the option for linear regression.

In the main regression menu, we will move over Sustainable Innovation Behavior as our
dependent variable and Cultural Intelligence (CQ), Knowledge Sharing (KS), and Organizational
Culture Differences (OCD) as our covariates (or predictor variables).

21 | P a g e
There is nothing we need to adjust in the “model” drop down, so we will move on to “statistics”.
In addition to the default settings, we will click:
● “Confidence intervals” under regression coefficients to get a range of likely slope values.
● “Statistics” under residuals to look at the typical amount of error in our model.
● “Descriptives” in the right column of options, just to look at typical values in our dataset.
● “Collinearity diagnostics” to continuing checking assumption six, that we do not have
multicollinearity.
● “Durbin-Watson” to check assumption seven, that we have independence of errors.

22 | P a g e
We will also skip over the model specification dropdown and move onto plots. In the plot
dropdown, we will click on Residuals vs. histogram to request residuals plotted as a histogram.

Now we’re ready to look at our output! First let’s begin by checking our remaining assumptions.
Assumption Six: Are our predictor variables highly correlated?
As mentioned previously, if our predictor variables are highly correlated with each other, this can
create problems in interpreting the results of the analysis. Collinearity is the term used to
describe this issue of high correlation between predictor variables. One way to check for
collinearity is by looking at the correlations between the variables. Another method is by
checking the Variance Inflation Factor (VIF) and Tolerance statistics. The VIF measures how
much of the variance in our model is increased due to collinearity (or correlation) between our
predictor variables. The Tolerance statistic is the reciprocal of the VIF and measures the
proportion of variation in one predictor variable that is not explained by the other predictor
variables in the model. In general, we want to see VIF values below 5 and Tolerance statistics
above 0.2. If the VIF values are higher than 5, it suggests that there may be too much collinearity
between the predictor variables, and this can lead to inaccurate estimates of the regression
coefficients. In other words, it can be difficult to determine which predictor variables are
contributing to the outcome variable. If our VIF values are above 5, we may need to: (1) remove
one or more predictor variables or (2) combine variables into composite scores.
Assumption Seven: Do we have independence of errors?
One important consideration in regression analysis is whether the residuals (the differences
between the actual values and predicted values) are correlated with each other or not. The
Durbin-Watson statistic is a measure of this correlation. When the Durbin-Watson statistic is
between 1.5 and 2.5, it suggests that there is not a significant correlation between the residuals or
errors in our model. This is important because if there is a high degree of correlation between
residuals, it can suggest that the model is not capturing all the relevant information about the

23 | P a g e
relationship between the variables. In other words, the model may not be a good fit for the data.
A Durbin-Watson statistic outside of the 1.5-2.5 range may indicate that there is a problem with
the model, and further investigation may be needed to determine the cause of the correlation
between residuals.
The Durbin-Watson statistic is a part of the Model Summary results. If this value is between 1.5-
2.5, we are not concerned there is a significant amount of correlation between the residuals.

Looking at the preceding figure, we can determine that our Durbin-Watson statistic is within the
acceptable range of 1.5-2.5. To report this in APA format:

The model has acceptable independence of errors, Durbin-Watson = 1.92.

Let’s look at our new output with three predictors. In the Model Summary table, we see that our
three variables together are correlated with Sustainable Innovation Behavior at .688 (Multiple R).
We also see that, together, Cultural Intelligence, Knowledge Sharing, and Organizational Culture
Differences explain 46.8% (R2) of the Sustainable Innovation Behavior. That number grew from
44.8% in our simple regression with only Cultural Intelligence. That tells us that Knowledge
Sharing and Organizational Culture Differences do help explain a little bit more about
Sustainable Innovation Behavior, but only about 2% more when we add all three predictors
together.

24 | P a g e
Our ANOVA table shows us that our linear model, including the three predictors, is explaining a
significant amount of variance in Sustainable Innovation Behavior. The F value of 99.36 has a p
value less than .001, which is below our .05 cutoff. This tells us that our overall model is
significant.

Now that we have determined that we have a significant overall model predicting Sustainable
Innovation Behavior in the workplace, we would like to gain a better understanding of the
relationship between each of our predictor variables with the dependent variable. To do this, we
are doing to look at the slopes of each of our predictors, which we’ll find in the coefficients
table. Our intercept now reflects if you are zero on all three predictors. So, we can say if a person
has no Cultural Intelligence, no Knowledge Sharing, and no Organizational Culture Differences,
the average Sustainable Innovation Behavior is 6.91. The slope for Cultural Intelligence has
gotten smaller now. That’s because Knowledge Sharing and Organizational Culture Differences
are explaining some of the variance now too, making the effect of Cultural Intelligence a bit
weaker. The slope of Cultural Intelligence accounting for Knowledge Sharing and
Organizational Culture Differences, is .294. Our associated t value of 10.187 has a p value of
<.001, which is less than .05. We would say that Cultural Intelligence is a significant predictor of
Sustainable Innovation Behavior when you also account for Knowledge Sharing and
Organizational Culture Differences. Knowledge Sharing does have a significant slope (b = .196, t
= 2.322, p < .02). For every one-unit increase in Knowledge Sharing, you can expect Sustainable
Innovation Behavior to increase by .196. Organizational Culture Differences does not have a
significant slope (b = .065, t = 1.881, p = .06). For every one-unit increase in Organizational
Culture Differences, you can expect Sustainable Innovation Behavior to increase by .065.

Our standardized slopes further show us that Cultural Intelligence (beta = .53) is a stronger
predictor than Knowledge Sharing (beta = .13) and Organizational Culture Differences (beta =
.10).

Coefficients table from multiple regression output in JASP.

25 | P a g e
The other piece of information we are interested in is the collinearity statistics in the last two
columns. The VIF measures how much of the variance in our model is increased due to
collinearity (or correlation) between our predictor variables. The Tolerance statistic is the
reciprocal of the VIF and measures the proportion of variation in one predictor variable that is
not explained by the other predictor variables in the model. In general, we want to see VIF
values below 5 and Tolerance statistics above 0.2. If the VIF values are higher than 5, it suggests
that there may be too much collinearity between the predictor variables, and this can lead to
inaccurate estimates of the regression coefficients. In other words, it can be difficult to determine
which predictor variables are contributing to the outcome variable. If our VIF values are above 5,
we may need to: (1) remove one or more predictor variables or (2) combine variables into
composite scores.
Our tolerance values range from .49 to .58, meaning that between 49% and 58% of the variance
in the predictors is unshared. Our largest VIF value is 2.06, which is below 10, so there is not
much concern for multicollinearity.

Related to our assumptions, we also want to make sure we take another look at our residuals plot
for our multiple regression model. Again, we see that our residuals for our model (based on the
regression equation with three predictors) approximate a fairly normal distribution.

26 | P a g e
Putting it all Together

It’s time to summarize the results of our multiple regression analyses.

Prior to conducting our multiple regression, we tested several assumptions. We examined our
predictors (Cultural Intelligence, Knowledge Sharing, and Organizational Culture Differences)
and outcome (Sustainable Innovation Behavior) variables for normality and any concerns with
outliers. All skewness and kurtosis values were all in the normal range. Upon inspection of
variable boxplots and histograms, four outliers were revealed that were retained for analysis.
Using a scatterplot, we also visualized the relationships between the variables. Upon viewing the
scatterplots, we observed that the relationships appeared linear and that there were no concerns
with heteroscedasticity. Plots of the residuals also showed that the residuals were approximately
normally distributed. We also looked at the correlation between our predictor variables, finding
that the variables were moderately correlated (r range = .52 to .63). The variance inflation factor
values for our model were all less than 2.05, providing support that there were not substantial
concerns with multicollinearity. The model has acceptable independence of errors, Durbin-
Watson = 1.92.

In our multiple regression, we examined how the Cultural Intelligence, Knowledge Sharing, and
Organization Culture affected Sustainable Innovation Behavior. We found that together, these
three predictors explained a significant amount of variance in Sustainable Innovation Behavior,
F (3, 336) = 99.3, p < .001. Together, the predictors explained 46.8% of the variance in
Sustainable Innovation Behavior. Organizational Culture Differences, Cultural Intelligence, and
Knowledge Sharing were significant predictors of Sustainable Innovation Behavior, when
controlling for the other predictors. Knowledge Sharing was positively related to Sustainable
Innovation Behavior (b = .20, SE = .08, t (332) = 2.32, p = .02). Cultural Intelligence was also
positively related to Sustainable Innovation Behavior (b = .29, SE = .03, t (332) = 10.19, p <
.001). Organizational Culture Differences was not a significant predictor of Sustainable
Innovation Behavior (b = .07, SE = .04, t (332) = 1.88, p = .06).

27 | P a g e
Real World Meaning
From this study, we could share that Cultural Intelligence, Knowledge Sharing, and Organization
Culture Differences together impact Sustainable Innovation Behavior, but the strongest
relationship exists between Sustainable Innovation Behavior and Cultural Intelligence, where the
more cultural intelligence someone has, the more likely they are to also engage in sustainable
innovation behaviors in the workplace. The same pattern is true with knowledge sharing, that
sharing more information is related to more innovation, just not as strongly as cultural
intelligence relates to innovation behaviors. Together, Cultural Intelligence, Knowledge Sharing,
and Organizational Culture Differences can explain over half of the variance in Sustainable
Innovation Behavior across a sample of individuals in organizations.

In their study, Li et al. (2021) found similar results. They examined how Cultural Intelligence
impacted Sustainable Innovation Behavior with Knowledge Sharing as an intermediary variable
and Organizational Culture Differences as a moderator. Their results showed that Cultural
Intelligence positively impacts Sustainable Innovation Behavior, but this relationship is mediated
or linked by Knowledge Sharing. They also found that Organizational Culture Differences did
not significantly moderate or change the relationship between Cultural Intelligence and
Sustainable Innovation Behavior, but it did impact the relationship between Cultural Intelligence
and Knowledge Sharing.
The results of our statistical analysis and that of Li et al. (2021) both suggest that Organizational
Culture Differences does not really have a significant impact on our main variable, Sustainable
Innovation Behavior. It can also be said that Sustainable Innovation Behavior and Cultural
Intelligence do have a strong relationship together whereas the relationship between Sustainable
Innovation Behavior and Knowledge Sharing is not as strong.
References
Gölgeci, I., Swiatowiec-Szczepanska, J., & Raczkowski, K. (2017). How does cultural
intelligence influence the relationships between potential and realised absorptive capacity and
innovativeness? Evidence from Poland. Technology Analysis & Strategic Management, 29(8),
857–871. https://doi.org/10.1080/09537325.2016.1245858
Hu, S., Gu, J., Liu, H. & Huang, Q. (2017). The moderating role of social media usage in the
relationship among multicultural experiences, cultural intelligence, and individual creativity.
Information Technology & People, 30(2), pp. 265-281. https://doi.org/10.1108/ITP-04-2016-
0099
Li, J., Wu, N., & Xiong, S. (2021). Sustainable innovation in the context of organizational
cultural diversity: The role of cultural intelligence and knowledge sharing. PloS One, 16(5),
e0250878–e0250878. https://doi.org/10.1371/journal.pone.0250878
Pandey, A. & Charoensukmongkol, P. (2019). Contribution of cultural intelligence to adaptive
selling and customer-oriented selling of salespeople at international trade shows: does cultural

28 | P a g e
similarity matter? Journal of Asia Business Studies, 13(1), pp. 79-96.
https://doi.org/10.1108/JABS-08-2017-0138
Authors
This guide was written and created by Ruth V. Walker, PhD, Ashlyn A. Moraine, and Kristen J.
Black, PhD.
Acknowledgements
We would like to thank Jinlong Li, Na Wu, and Shengxu Xiong for generously publishing their
data through PloS One.
Copyright

CC BY-NC-ND: This license allows reusers to copy and distribute the material in any medium
or format in unadapted form only, for noncommercial purposes only, and only so long as
attribution is given to the creator.

29 | P a g e

You might also like