This chapter will introduce the methods and instruments that will be used to conduct this study. This chapter will also include the research framework, procedure, design, data collection and measurements.
Research Framework
Figure 3.1 Research Framework. This showed the relationship between the variables in this study. There are two independent variables, perceived surface-level (including age, gender and race/ethnicity) and perceived deep-level diversity (values) and one dependent variable, team social integration (including team cohesion).
All the variables mentioned above have all been proposed and reviewed in the literature review in chapter two. In this study, the researcher followed previously used instruments which have been tested and proven to be valid and reliable.
Figure. 3.1. Research framework
Perceived Surface-level Diversity
Perceived Deep-level Diversity
Team Social
Integration
19
Research Hypothesis
Hypothesis 1. Perceived Surface-level diversity will affect team social integration Hypothesis 2. Perceived Deep-level diversity will affect team social integration
Research Procedure
This study has 10 steps in the research procedure to be conducted. This section shows the figure of research procedure and explains the steps of the procedure. In first step the, researcher identified the population or sample interested and the problem that is occurring in the society. Next the researcher examined and reviewed the literature to find out what are the variables and elements which claim to affect the problem positively. After reviewing the literature and identifying out the variables, the next step was be to develop the research topic look and develop the purpose and the importance of the research.
The research framework and hypothesis were developed based on the literature review and modified where needed. The instruments used to measure the variables were identified through the literature used in the study. None of the measurements were modified but will be modified if necessary.
A pilot test was done to measure the validity and reliability of the measurements.
After the pilot test the instruments was reviewed and modified where necessary. Next, we collected data for the main study and ran the analysis for the data collected. Last, we gave the conclusion and some suggestions for future studies. Below is the figure 3.2 will show the Research Procedure.
20
Step 1: Identify problem and population interested
Figure 3.2. Research procedure
Step 2: Review of literature
Step 3: Develop research topic and purpose
Step 4: Develop research framework and hypothesis
Step 5: Develop research instruments
Step 6: Conduct pilot test
Step 7: Revise instrument
Step 8: Collect data
Step 9: Perform data analysis
Step 10: Develop conclusion and suggestions
21
Research Design
This section talks about the overall design of the study. This study carried out a quantitative research by means of survey questionnaire. This was used to collect the data on measures of perceived surface-level and perceived deep-level diversity and team social integration. Participants filled out a questionnaire assessing their perceived level of the different diversity and team social integration. Statistical analysis tools was carried out to arrive at the study's findings and conclusion.
Sampling and Data Collection
A cross-sectional design was used by collecting data from participants from schools in Taiwan. The participants that were investigated are students, to be more specific, undergraduate and graduate students. The students were from mainly northern Taiwan (e.g.
Taipei city, New Taipei City, Taoyuan, Keelung, Hsinchu) and some from other cities around Taiwan. The participants were expected to have previous team or group experience in school from team assignments or group projects.
The participants could have been in any discipline in any department, as long as they had previous team experience which was not less than one month and at least three members in the team was the acceptable criteria for the study. The study targeted students' perception on each individual in their team/group based on items in the survey questionnaire.
This study used convenient sampling. Data were collected from internet and social media as well as hard copy questionnaires. Online and social media included Face Book and Google surveys.
22
Sample Profile
For the full study, a total of 227 questionnaires were returned (175 online, 52 hardcopy). After removing questionnaires because of missing sections and not meeting the criteria, if participants had less than 1 month experience in a group they did not meet the criteria, a total of 207 questionnaires were useable for the study, 18 respondents had less than 1 month experience and 2 had missing sections. The demographics of the participants are described below. The majority of people who answered were between 18 and 27 years of age, that is 77.8% of the total participants, also there were 2 missing data for age. As for gender there were almost double the amount of females to male who responded, male 35.7%, female 64.7%. The majority of people were of Asian background 81%. As for the number of people in each group, there was quite a spread of number, 3-person group = 4.3%, 4-person group =19.3%, 5-person group = 16.9%, 6-person group = 23.2 % and 6(+)- person group = 36.2 %. As for the class type, there were four types which stood out among the rest, Freshmen = 20.3%, Seniors = 16.4%, first years = 21.2% and Master-second years = 27.5%. The majority of people spent more than 4 months in a particular group at 65.2% and the majority of people had many experiences being in a team or group environment at 56.5%. There were students from 57 majors because of so many majors, only half of the percentage were included which included MBA's, Banking/Finance/Statistics, applied English, engineering, tourism and management and information management. A full summary of the sample characteristics is shown below in Table 3.1.
23 Table 3.1.
Descriptive Statistics on Sample Characteristics (N=207)
Item Category Frequency Percentage (%)
Age 0 2 1
Race/Ethnicity Asian 168 81.2
North American 3 1.4
Master-Second Year 57 27.5
Master-Third Year and
24 Table 3.1. (Continued)
Item Category Frequency Percentage (%)
Major MBA 40 19.3
Banking/Finance/Statistics 20 9.7
Applied English 12 5.8
Engineering 12 5.8
Tourism and Management 11 5.3
Information Management 10 4.8
Method of Data Analysis Descriptive Statistics
Since this study did not use random sampling method, descriptive statistics helped to calculate the important features of the data such as sample mean, mode, standard deviation (SD), demographics and others. Descriptive statistics was very helpful for the study as it helped to summarize and describe the important features of the data collected.
Factor Analysis
The study conducted Exploratory Factory Analysis (EFA) in SPSS and Confirmatory Factor Analysis (CFA) in AMOS for construct validity of the measurements. EFA was used to differentiate properties in the data collected. It was also used to check for common method variance (CMV). Harmon's single factor test is observed in EFA, this was used to check for common method variance (Podsakoff & Organ, 1986). CMV comes about when two or more measures derive from the same source. Some recommendations to minimize CMV include reordering scale, while varying the measurement scales. On the other hand CFA is a statistical tool used to observe relationships among latent variables (Jackson, Gillaspy Jr., & Purc-Stephenson, 2009) and evaluates the study's hypothesis. To utilize CFA the researcher was required in advance to hypothesize a number of factors and identify whether or not those factors were correlated. In sum, CFA tests whether the data collected matches the theorized measurement model of a variable.
Correlation Analysis
In this study, Person's Correlation analysis was used to test the relationship between variables. This method results in correlation coefficients which enables the observation of
25
how significant one variable is related to another variable. This allows for the observation of positive significance or negative significance and strengths between the variables.
Hierarchical Regression Analysis
Hierarchical regression is a tool used for analysis when variance on a criterion variable is being explained by predictor variables that are correlated with each other (Lewis, 2007). So hierarchical regression was used to test the hypothesis in the study and check if there are other predictors which may affect the dependent variable.
Structural Equation Modeling (SEM)
Structural equation modeling or SEM is a statistical procedure that is used to test and estimate causal relationships (Jackson et al., 2009). This technique uses both causal assumptions and statistical data.
The data that was collected for this study was analyzed using AMOS, which is a type of SEM in SPSS. The difference with AMOS is that it automatically includes estimation of variances for all your independent factors. It also allows for CFA and creating path diagrams (Byrne, 2013). AMOS uses the maximum likelihood estimation as the default feature (Byrne, 2001).
Deep-level diversity and Team social integration were ran independently in AMOS for CFA. The outputs which had the highest values for interest included χ2/df, RMR, GFI, AGFI, RMSEA, CR and AVE to examine the measurement models goodness of fit. The χ2/df refers to the chi-square divided by degrees of freedom. RMR is the root mean square residual which is an indicator of how much estimated covariance and variances are different from the observed ones. GFI or goodness of fit is a value that analyzes the proportion of variance that is accounted for by the projected population covariance. AGFI or the adjusted goodness of fit is the index of adjusted GFI. RMSEA or root mean square error of approximation compares the lack of fit to the saturated model (Hooper, Coughlan, &
Mullen, 2008). Below is Table 3.2 which shows the criteria and acceptable cutoffs for the model fit. Surface-level diversity was not ran in SPSS AMOS as it is a non-latent variable.
26 Table 3.2
Index of Model Fits
Good fit Acceptable fit Author's notes
χ2/df 2-5
Note. Summary based on Hooper, Coughlan, and Mullen (2008) (top rows) and Schermelleh-Engel, Moosbrugger, and Müller (2003) (bottom rows). Adopted from Cleveland (2015).
Measurements
The questionnaire used in this study used different measurements for each variable.
The next section goes into detail each instrument of measurement that was either adopted or adapted. Below are the descriptions of each of the measurement and written in order as presented in the full scale study.
Team Social Integration (TSI)
For the dependent variable it was adopted using Carron, Widmeyer and Brawley's (1985) team cohesion questionnaire which included 9 questions. The statements in the questionnaire asked respondents to rate their level of cohesion and satisfaction in their team experience. An example statement is, "our team is united in trying to reach goals for performance". Items were scored on a 7-point Likert scale (1 = Strongly Disagree; 7=
Strongly Agree). The internal consistency of the measure was acceptable ranging from .63 to .81 (Carron, Widmeyer & Brawley, 1985).
Perceived Surface-level Diversity
Perceived Surface-level diversity measures was adopted using Chang's (2009) measurements. This measurement asked respondents to rate their level of difference in terms
27
of age, gender and race/ethnicity in their team experience. An example statement is, " How diverse do you think your team members are to you in terms of age". If , for example, the group/team had a total of six members including the rater, the rater will answer this five times with his/her perception of that team member in relation to their age/gender/race or ethnicity, the responses would then be averaged to have an overall individual-level measure of perceived variability (Chang, 2009). The items were scored on a 5 point Likert scale (1=
not at all different; 5= very different).
Perceived Deep-level Diversity
The index of perceived deep-level diversity was adopted from Chang (2009) to operationalize diversity. The measurement asked respondents to report on a 6-point scale ranging from very much like them to not like them at all scale regarding how similar they thought they were on average to their group members (1 = very much like them; 6 = not like them at all). They were given 5 statements and each statement described a person in which they had to report how similar they were to their group member. An example statement is "It is important to them to be in charge and tell others what to do. They want people to do what they say". The internal consistency for the measurement was high in previous studies and accepted. Similar to perceived-surface level indices, the rater would indicate based on their perspective how different the individual is compared with themselves.
Control Variables
There were nine demographic questions in total and out of the nine, three questions were used as control variables. Team size according to Harrison et al., (2002), Mohammed
& Angell, 2004), Newell et al. (2008), Liao et al. (2008) and Chang (2009) is used as a control variable in this study. A larger group may have more potential for more heterogeneity and size may influence cohesion and performance. Mohammed and Angell (2004) examined length of time in a team and times been in a team as control variables as the sample included undergraduate and graduate students. These two variables are also included in this study as control variables
Validity and Reliability
Construct validity and content validity were used to measure validity in the study.
Content validity refers to the extent at which a questionnaire reflects the indented domain of content, which means the instruments measure what they are meant to measure. Content validity was established for all measurements while the instruments were either adopted or
28
adapted from previous studies. A pilot test with a sample of 42 was conducted to check the validity and reliability of the measurements for the main study. This helped to improve the study's design before the main research was carried out. While doing the pilot test and the main study, exploratory factor analysis (EFA) was used to examine the factor structure and the threat of common method variance (CMV). Construct validity was also carried out in the main study using confirmatory factor analysis (CFA). The results of both EFA and CFA are reported below.
Reliability analysis was also run for both the pilot and the main study. Cronbach's alpha reliability test was used to measure the internal consistency of the measurements.
Exploratory Factor Analysis
Exploratory factor analysis was conducted to observe the factor and cross factor loadings of the measurement items in the study. Using EFA, a Harmon's single factor test was done on the items to observe common method variance for each survey which was answered by individual sources, the unrotated variance accounted for was 24.17% which is below 50% as suggested by Podsakoff et al. (2003). In the pilot test, Team social integration and Deep-level diversity were ran through EFA and there were 4 factors extracted with an eigenvalue larger than 1 and a cumulative variance of 72.05%, the Kaiser-Meyer Olkin (KMO) was used to measure the adequacy of the sample and it had a value of 0.678 and the Barlett's test of Sphericity was .000 which is significant. In the pilot test the deep-level variables were grouped together so they measured the component well meaning that the items explained the measurement well and were not cross-loading into other variable items and team social integration was not grouped as well as deep-level diversity, which was an indication that the respondents saw team social integration as more as a multi-dimensional construct, below in table 3.3 and 3.4 is a table showing the Kaiser-Meyer Olkin and the rotated factor loadings respectively.
Table 3.3.
KMO and Bartlett's Test
Kaiser-Meyer-Olkin Measure of Sampling Adequacy. .678
Bartlett's Test of Sphericity
Approx. Chi-Square 313.448
df 91
Sig. .000
29 Table 3.4.
Exploratory Factor Analysis (EFA):Team Social Integration and Deep-level Diversity(Pilot Test)
Team social integration 9 .867
Team social integration 8 .812
Team social integration (R) 6 .651
Team social integration (R) 7 .838
Team social integration 5 .651
Team social integration (R) 3 .628
Team social integration (R) 2 .515
Team social integration 1 .907
Team social integration (R)4 .614
Note. Extraction Method: Principal Component Analysis.
Note. (R) = A Reverse Question.
Confirmatory Factor Analysis
Confirmatory factor analysis was ran on each set of items separately through AMOS.
The results determined whether any modification is needed to improve the model fit. The purpose of CFA is to find if the data fit the theoretical model measurement. After running the items through CFA, it was then examined whether which items should be deleted to meet the criteria. After the removing of items the modified list of items were ran again through CFA. If the new list met the criteria of the model fit, it would normally be ran against another collection of data to cross validate. Though, items in this research could not undergo this procedure because of time and resource constraints. Nonetheless, the study's data sample was randomly separated into two groups and cross validated in this method as suggested by other researchers (e.g. Blunch, 2013; Browne, 2000), which yielded a sample comparison of model fit output.
CFA for team social integration. The results from the EFA showed that 3 factors were classified into 3 dimensions for Team social integration. The data was categorized and sorted in CFA based on the classifications and dimensions from the pilot test EFA. In CFA, factor 1 had questions TSI 9, TSI 8 and TSI(R)6, factor 2 included questions TSI(R) 7,
30
TSI(R) 2, TSI 5 and TSI(R) 3 and factor 3 included questions TSI 1 and TSI(R) 4. Out of the nine items, two were deleted by examining the modification indices (M.I) in CFA. TSI 5 question had high M.I with different questions so the model fit would improve if it is deleted.
TSI 9 also had high M.I with other questions and would improve the model fit if deleted. A summary of the original and modified model fit is shown below in table 3.5 and figure 3.3 shows the standardized regression weights for the modified items and other important values.
Figure 3.3. Team social integration CFA measurement model.
Table 3.5.
Team Social Integration Model Fit Summary
TSI χ2 df P χ2/df RMR GFI AGFI RMSEA AVE CR
Full items (list 9)
286.431 27 .000 10.61 .18 .77 .61 .22 .40 .83
Modifi ed items (7)
59.345 14 .000 4.24 .07 .92 .84 .13 .50 .87
31
After removing the two questions the model fit improved and all the other criteria improved such as GFI (.921), AGFI (.842), χ2/df (4.24). Since the measurement model was modified based on data instead of a theory, the researcher conducted a cross-validation of the modified measurement model. The cross-validation was achieved by splitting the sample randomly into two groups and performing a multi-group comparison on the modified team social integration measurement model. The result of the multi-group comparison of team social integration items was shown in table 3.4. The p-values show insignificances indicating constraining the two randomly split samples as equal does not significantly worsen the model fit. The modified measurement model was successfully cross-validated with two samples. To see a list of the items of team social integration which were deleted, refer to table 3.5
Table 3.6.
Multi Group Comparison for Cross Validation of Measurement Model TSI
Model DF CMIN P NFI
TSI 1 Our team is united in trying to reach its goals for performance
.36***
TSI 2 I’m unhappy with my team’s level of commitment to the task
- TSI 3 Our team members have conflicting aspirations for the
team’s performance
.34***
TSI 4 This team does not give me enough opportunities to improve my personal performance
.76***
TSI 5 Our team would like to spend time together outside of work hours
Deleted TSI 6 Members of our team do not stick together outside of
work time
- (Continued)
32 Table3.7. (Continued)
Item Description Status SDW
TSI 7 Members of our team would rather go out on their own than get together as a team
.73***
TSI 8 For me this team is one of the most important social groups to which I belong.
.25**
TSI 9 Some of my best friends are in this team. Deleted Note. SDW = Standardized Regression Weights
CFA for perceived deep-level diversity. CFA of the 5 items of perceived deep-level diversity yielded acceptable indices for the goodness-of fit index; χ2/df (15.09), RMR( .04), GFI(.97), AGFI(.92), RMSEA(.09), AVE(.43) and CR(.73). Below is figure 3.4 showing the perceived level variable measurement model and table 3.8 showing perceived deep-level diversity model fit summary. Also table 3.9 shows the perceived deep-deep-level variable
CFA for perceived deep-level diversity. CFA of the 5 items of perceived deep-level diversity yielded acceptable indices for the goodness-of fit index; χ2/df (15.09), RMR( .04), GFI(.97), AGFI(.92), RMSEA(.09), AVE(.43) and CR(.73). Below is figure 3.4 showing the perceived level variable measurement model and table 3.8 showing perceived deep-level diversity model fit summary. Also table 3.9 shows the perceived deep-deep-level variable