Results and Analysis

9.1

Introduction

The data were analysed in two parts. In Part I, the data were prepared, checked for norm ality and reliabilities and validity for the measures used in this study were established. The data were analysed using SPSS version 12 and M icrosoft Excel. In Part II the main model, minor models and their concomitant hypotheses were tested.

Pa r tI: Da t a Pr e p a r a t i o n

9.2

Data Preparation and Preliminary Analyses

The data were gathered from the email using a custom Java batch processing application and converted to a M icrosoft Excel file. Prelim inary data preparation and calculations were then undertaken. Firstly, the data were aggregated to team level. Follow ing this, both individual level and team level data were transferred to SPSS version 12 for further analysis. The internal consistency o f the measures was assessed along with the level of team agreement. The data were then checked for normality. After this initial data preparation, the data were described at team level in terms o f means and standard deviations.

Inferential analyses were then carried out. A series o f independent t-tests were performed to assess the differences, if any, between Ireland and the U K on responses on all measures.

9.2.1 Data Preparation

Norm ality was checked on all variables by assessing skewness and visually inspecting histograms. Using the conservative convention that the skewness statistic is not more than twice as large as its standard error (Tabachnik & F id cll, 2001) then the data were not reliably different from normal, for all variables except team size, fam iliarity with written procedures, and presence o f expertise. Further analyses o f these variables indicated that skewness was attributed to the presence o f outliers. To check if these outliers posed a threat to the inferential analysis, the regression analyses were conducted

with and without the outliers, which were found to exert no significant influence. It was decided to keep the outliers in the data. In addition, the data for these regressions were linear since the standard deviation for the dependent was more than the standard deviation for the residuals. In regression analysis in particular, skewness is only an issue if it relates to the dispersion o f the residuals. For all regression analyses there were no outlying residuals i.e. points whose standardised residual is greater than 3.

The data met the conditions for regression. Inspection o f standardised residual plots indicated homosecdasticity, where the residuals were dispersed random ly throughout the range o f the estimated dependent. The correlations were all w ell below the 0.8 that would indicate high collinearity (Kennedy, 1985). In addition, farther evidence to indicate that m ulticollinearity was not a problem, was apparent in the variance inflation factor (V IF ) w hich was below 4.0 (w hich is the recommended cut-off point; M iles & Shevlin, 2001) for all variables in all regressions.

9.2.2 Transform ing and Weighting Scores

In order to maintain consistency across variables, each score on the T T K M was subtracted from the maximum score, so that a higher the score on T T K M reflected more expert like responses. In addition, ‘reliance on gut instinct’ and ‘reliance on written procedures’ were subtracted from the m axim um +1, in order that high scores were equal to high reliance.

In relation to the transactive memory system (TM S) measure, a weighted composite score was computed on the advice o f Lew is the originator o f the measure (personal communication, 2004). The technique, described more fu lly in K im and M ueller (1978), essentially involves regressing the TM S factor on its sub-factors and items, w hile still taking into account the hypothesised measurement model. Scale weights are given by the regression coefficients. In this study the scale weights were as follows: specialisation: R 2 = 0.53, credibility: R 2 = 0.79, coordination^2 = 0 .67. The scores for each sub-factor were m ultiplied by their scale weight the three were added together to make the weighted composite.

9.3

Aggregation of Data across Countries and to Team Level

Scores on all measures were aggregated to team level. First, individual scores were calculated for most variables (except proxim ity, diversity, team size and new product development capability, w hich were collected at the team level), then, these were averaged for team level analysis. Criteria for scoring and averaging the test scores were outlined in Chapter 7.

A series o f t-tests on the main variables indicated no significant differences between Ireland and the U K on any o f the variables in this study (Appendix M , Table M .l). Country o f origin was not an influencing factor in the present study, therefore the organisations were pooled into a single sample form ing a combined sample size o f 48 teams.

9.4

Adequacy o f Measures

A ll constructs using m ultiple indicators were tested for their reliability at both the individual and team level. The validity o f all measures, where relevant was established using factor analysis. However, the T T K M was tested for convergent and discrim inant validity using Pearson’s correlation coefficient.

The means and standard deviations for all measures are presented in Table 9.1, along with the Cronbach alpha reliabilities, at both the individual and team level. In addition, the reliabilities for the measures at the team level from the previous studies discussed in Chapter 8 are also presented. Finally, the two values for inter-rater agreement are reported: rwg (James et al.1984) and R wg (Lin d ell et al. 1999).

9.4.1 Internal Consistency of Measurements

The internal consistency for all measures at individual level are all above a = 0.68 (except for the T T K M ) and above a = 0.67 at the team level. Therefore, the internal consistency o f the measures is considered adequate at the team level. Internal consistency for the T T K M , as measured by Cronbach’s coefficient alpha, was a =0.49 at the individual level and a = 0.71 at the team level, indicating a significant increase in the internal reliab ility o f the measure at the team level, thus providing support for the premise that T T K M measures tacit knowledge at the team rather than individual level. Given that the obtained team level reliability falls w ithin the range for other situational judgement tests and for those reliabilities obtained on previous measures o f tacit

knowledge (Hedlund et al. 2003). Then the internal consistency o f the team level score is considered to be acceptable.

Table 9.1 Means, Standard Deviations, Reliability and Inter-Rater Agreement for All Measures

Variable Mean SD Min Max alpha

Ind. alpha Team alpha Prev. rwg R Wg 1. Quality o f SI 12.83 1.88 9.17 16.00 NA NA NA NA N A 2. Quantity o f SI 64.60 17.06 1 0 . 0 0 1 0 0 . 0 0 N A NA NA NA NA 3. TMS (weighted 41.94 4.38 31.83 47.76 . 8 8 .92 .82 .98 .74 composite) 4. Credibility 21.64 2.34 15.00 25.00 .82 . 8 8 .79 .96 .75 5. Specialisation 20.85 3.04 13.00 25.00 . 8 8 .94 .76 .95 .72 6. Coordination 20.59 2.91 14.00 25.00 . 8 8 .94 .82 .95 .73 7. Team Tacit 5.49 2.48 0 . 0 0 10.08 .49 .71 NA .96 .58 Knowledge (10.84) (6.26) (16.34) 8. Effectiveness 3.69 0.55 2.40 4.60 .76 . 8 8 . 8 6 .90 .57 9. Efficiency 3.24 0.73 1.50 4.50 . 6 8 .83 .74 .76 .59 10. Psychological 5.62 0.64 4.31 6.57 .78 .82 .82 .93 .60 Safety 11. Team size 4.91 2.34 2 . 0 0 1 1 . 2 0 NA NA NA NA NA 12. Diversity 50.17 38.07 0 . 0 0 1 0 0 . 0 0 N A NA NA N A NA 13. Formal knowledge 9.16 1.93 4.00 14.00 .78 .85 NA .87 .40 Sharing System 14.Experience (years) 11.64 4.97 2 . 0 0 22.50 NA NA NA NA NA 15. Expertise 80.00 13.11 33.33 98.33 . 6 8 .67 . 8 8 NA NA Presence 16. Reliance on Gut 4.00 0.62 3.00 5.00 NA NA NA NA NA Instinct | (2.0 0) 17. Familiarity with 4.07 0.76 1 . 0 0 5.00 NA NA NA NA NA Written Procedures! 18. Reliance on 3.01 0.95 1 . 0 0 5.00 NA NA NA NA NA Written Procedures! (3.00) 19. Administrative 3.13 0.72 1 . 0 0 4.50 .81 .89 .82 .80 .41 Coordination 20. New Product 21.29 2.58 16.00 27.00 NA . 6 8 NA NA NA Development

() items in brackets are non-reverse coded.

The 5-item scale for ‘new product development capability’ initially had low reliability (a = 0.54), two items were removed increasing the reliab ility to a = 0.69. The items relating to ‘ability to price com petitively’ and ‘ ability to penetrate new markets’ were removed. Leaving three usable items for analysis in the scale, consisting o f items a, b and c (see Appendix C.9).

The first set o f inter-rater agreement values or rwg; are the values reported in a ll studies using the formula forwarded by James et al. (1984). The rwg values range from 0.76 to 0.96 and thus reflect a high level o f within-team agreement. Lindell et a l.’ s (1999) measure incorporates weighting based on sample size, allow ing for inter-group comparison o f inter-rater agreement (R wg). The inter-rater agreement is significantly lower when this form ula is applied, ranging from 0.40 to 0.75. This is because the scale range is from -1 to +1 for 5-item scales and from -3 .5 to +1 for 7-item scales. Comparison between the 5 and 7 item scales can be made because all were positive. However, comparison cannot be made between James et al. (1984) and Lindell et al. (1999), since the form ula are different. Hence, both inter-rater agreement scores are reported.

9.4.2 Statistical Validity of the Measures

Validation o f the T T K M was established using M essick’s (1995) framework, w hich was discussed in Chapter 8. Theoretical and statistical validities for all other scales had already been established (see Chapter 7) except for presence o f a formal knowledge sharing system and new product development capability. Principal components analysis with varim ax rotation, a cut-off criterion o f .40 for factor loadings and eigen values o f above 1.0 or above, was used to check the existing statistical validities, and to establish the validity o f ‘ formal knowledge sharing system’ , and ‘new product development capability’ . These were all conducted on team level data.

The three items that make up the ‘formal knowledge sharing system’ loaded on one factor. These items were entered together with administrative coordination and two clear factors were revealed, providing discrim inant validity for presence o f a formal knowledge sharing system (see Table 9.2). The three items that constitute ‘new product development capability’ were entered into the analysis with the measures o f efficiency and effectiveness and three distinct factors were elicited. Thus the two factors that

constitute team performance were confirmed and discrim inated from one another and from new product development capability. The results for these analyses are illustrated in Table 9.3.

Table 9.2 Factor Analysis for Formal Knowledge Sharing system and Administrative Coordination

Items Adm inistrative K nowledge

Requirements/design review . 8 8

Project milestones and delivery schedules .87

Project documents and memos .85

Regularly scheduled team meetings .74

Formal policies and procedures for coordinating teams work .74

Design inspections .70