Hypotheses testing

HLM analysis included there models: null model, random-coefficient regression model, and intercepts- and slopes -as-outcomes model.

Furthermore, HLM cannot be performed with missing data, so 22 data are removed. This study uses 194 valid data to examine hypotheses.

4.1.3.1 Regression analysis

This study conducted regression analysis to test H1 to H3, and to understand the influences on NFC, openness, and emotion on learning behavior.

According to the results in Table 4-3, only openness did not significantly affect learning behavior. Therefore, H1 and H3 were supported, but H2 was not supported. Based on these results, openness did not have direct effects on learning behavior, so the moderating effects of openness (H9, H12, H15, and H18) were not analyzed.

Table 4-3 Regression analysis

This study examined interclass correlation coefficient (ICC [1]), reliability of the mean (ICC [2]), and reliability with group interrater agreement (rwg) before perform HLM. The equations were as following:

ICC [1] = ₂

where τ00 was variances between-group; σ² was variances within-group,

49

and n was samples.

The result showed τ00 = 0.098, σ² = 0.780, thus, ICC [1] = ρ = 0.1116. This means 11.16 percent of variance in learning behavior resided between the technology companies, 88.84 percent of the

variance resided within the technology companies. Following the above equations, this study computed ICC [1] and ICC [2], and the values were .11 and .96 for learning behavior; .18 and .98 for interpersonal relationship; .20 and .98 for social relationship; .21 and .98 for collaborative culture; .20 and .98 for organizational climate.

rwg(j) = the expected variance of a hypothesized null distribution (Castro,

2002).

Then, this study checked rwg(j) values of organizational level:

interpersonal relationship, social relationship, organizational

collaborative culture, and organizational climate, and their rwg(j) values had to exceed .70. This study computed rwg(j) values of the four

variances, and obtained the median rwg(j) values of interpersonal relationship was .93, social relationship, .95; organizational

collaborative culture, .95; organizational climate, .94. All these r_wg(j) values are above conventionally acceptable value of .70 (Liao &

Chuang, 2004). Thus, this study concluded aggregation was justified for these variables.

1. Null model

In this part, this study examined the effects of cross-levels, the two level models were:

50

Level 1: Yij = β0j + rij

Level 2: β0 j = γ00 + u0j

Mixed: Yij = γ00 + u0j + rij

where Yij was learning behavior, i means workers, j means

technology companies, β0j was the intercept at Level 1, rij was the error at Level 1, γ00 was the intercept at Level 2, u0 j was the error at Level 2.

The hypotheses predicted that both individual- and organizational-level variables significantly affected learning

behavior. As earlier state, ICC [1] = 0.1113 meant that HLM was an acceptable statistical method in this study. Additionally, γ00 in final estimation of fixed effects with robust standard errors should be checked. Through null model, γ00 = 5.685 (t = 61.203, p = .000).

Thus, random-coefficient regression model could be tested. The result of null model was shown in Table 4-4.

Table 4-4 Null model

Fixed effect Coefficient Standard error t Intercept 1, β0

Within-companies, σ² 0.883 0.780

Deviance = 517.365

Note: ^***p < .001; ^† p < .10

2. Random-coefficient regression model

This study put NFC and emotion in the model, and obtained:

Level 1: Y_ij = β_0j + β_1j X_1ij + β_2j X_2ij+ rij

Level 2: β0 j = γ00 + u0j

β

1 j = γ10 + u1j

β

_{2 j} = γ₂₀ + u_2j

51

Mixed: Yij = γ00 + γ10X1ij + γ20X2ij + u0j + u1jX1ij + u2jX2ij +rij

where Yij was learning behavior, X1 ij was NFC, and X2 ij was

emotion, β0j was the intercept at Level 1, β1j and β2j were the slopes of NFC and emotion, γ₀₀ was the intercept at Level 2, γ₁₀ and γ₂₀ were the intercepts of NFC and emotion, rij was the error at Level 1,

u

_{0 j} was the error at Level 2, u_1j and u_2j were the errors of NFC and emotion.

This study examined H1 and H3 of this study through performing this model, and γ10 was the parameter estimation between the independent and dependent variables at Level 1. In Table 4-4, γ10 = 0.342 (t = 2.426, df = 42, p = .020), γ20 = 0.247 (t = 1.764, df = 40, p = .084), the results were significant and H1 and H3 were supported. Thus, knowledge sharing positively affected

learning behavior. Regarding to variance components, τ00 = 0.107, df

= 42, χ² = 52.228, p = .000, this result was significant, and it referred that 43 technology companies had significantly difference for NFC and emotion. Hence, the third model could be performed.

Table 4-5 Random-coefficient regression model

Fixed effect Coefficient Standard error t Intercept 1, β0

Within-companies, σ² 0.786 0.617

Deviance = 487.209

Note: ^***p < .001 ^* p < .05 ^† p < .10

52

3. Intercepts and slopes -as-outcomes model

The first step in this model was to examine intercepts which could be explained by the organizational environment factors, and the models were:

Level 1: Y_ij = β_0j + β_1jX_1ij + β_2jX_2ij + rij

Level 2: β0 j = γ00 + γ01Z1j + γ02Z2j + γ03Z3j + γ04Z4j + u0j

β

1 j = γ10 + γ11Z1j + γ12Z2j + γ13Z3j + γ14Z4j + u1j

β

2 j = γ20 + γ21Z1j + γ22Z2j + γ23Z3j + γ24Z4j + u2j

Mixed: Yij = γ00 + γ01Z1j + γ02Z2j + γ03Z3j + γ04Z4j +

γ

10 X1ij+ γ11X1ijZ1j +

γ

₁₂X_1ijZ_2j + γ₁₃X_1ijZ_3j + γ₁₄X_1ijZ_4j + γ₂₀X_2ij+ γ₂₁X_2ijZ_1j + γ₂₂X_2ijZ_2j +

γ

23X2ijZ3j + γ₂₄X2ijZ4j + u_0j + u_1jX1ij +

u

2j X2ij+ rij

where Yij was learning behavior, X1 was NFC, X2 was emotion, and Z1 to Z4 represented interpersonal, social interpersonal, collaboration, and organizational climate, γ00 was the intercept at Level 2, γ01 to

γ

04were the slopes for interpersonal, social interpersonal,

collaboration, and organizational climate, γ10 and γ20 were the intercepts of NFC and emotion, γ11 to γ14 and γ21 to γ24 were the interaction coefficients, rij was the error at Level 1, u0 j was the error at Level 2, u1j and u2j were the errors of NFC and emotion.

This model was used to examine H4 to H7, and to check γ01, γ02,

γ

03, γ04.

Table 4-6 presented the following results: γ01 = - 0.135 (t = -0.981, df = 38, p = .333), so this result was not significant and H4 was not supported; γ02 = -0. 005 (t = -0.027, df = 38, p = .979), so this result was not significant and H5 was not supported; γ03 = 0. 371 (t = 3. 002, df = 38, p = .005), so this result was significant and H6 was supported; γ04 = 0. 233 (t = 1.973, df = 38, p = .055), so this result was significant and H7 was supported.

53

In order to understand the moderation effects of interpersonal, social interpersonal, collaboration, and organizational climate on the influence of NFC in learning behavior, H8, H11, H14, and H17 by the values of γ₁₁, γ₁₂, γ₁₃, and γ₁₄ should be checked. Table 4-6 showed γ₁₁ = -1.113 (t = 3. 752, df = 38, p = .000), γ₁₂ = -1.109 (t = -4.278, df = 38, p = .004), γ₁₃ = 1.177 (t = 2.242, df = 38, p = .031), and γ₁₄ = -1.334 (t = -2. 664, df = 38, p = .012). Therefore, H8, H11, H14 and H17 were supported.

For the moderation effects of interpersonal, social interpersonal, collaboration, and organizational climate on the influence of

emotion in learning behavior, then examined H10, H13, H16, and H19 by the values of γ21, γ22, γ23, and γ24. In Table 4-6, γ21 = 1.163 (t

= 2.913, df = 38, p = .006), γ22 = 0.254 (t = 0.470, df = 38, p = .640),

γ

23 = -0.504 (t = -1.262, df = 38, p = .215), and γ24 = 0.486 (t = 1.354, df = 38, p = .184). So H10 was supported, but H13, H16, and H19 were not supported.

According to the results of the above three models of HLM, the researcher provides the full model, and Table 4-7shows the results.

54

Table 4-6 Intercepts and slopes -as-outcomes model

Fixed effect Coefficient Standard error t Intercept 1, β0

Within-companies, σ² 0.764 0.584

Deviance = 464.656

Note: ^***p < .001 ^** p < .01 ^* p < .05 ^† p < .10

55 Table 4-7 Hierarchical linear modeling results

Null model Random coefficient regression model Intercepts and slopes-as-outcomes model Level 1

Interception 5.685^*** (0.093) 5.702^*** (0.088) 5.666^*** (0.072)

NFC 0.342^* (0.141) 0.383^** (0.102)

Emotion 0.205^† (0.116) 0.154^† (0.089)

Level 2

Interpersonal

relationship -0.135 (0.138)

Social relationship -0.005 (0.166)

Collaborative culture

0.371^** (0.124) Organizational

climate 0.233^† (0.118)

IR×NFC -1.113^*** (0.260)

SR×NFC -1.019^** (0.325)

CC×NFC 1.177^* (0.525)

OC×NFC -1.334^* (0.501)

IR× E 1.163^** (0.399)

SR× E 0.254 (0.540)

CC× E -0.504 (0.399)

OC× E 0.486 (0.359)

Level 1 residual 0.780 0.617 0.584

Deviance 517.365 487.209 464.656

Note: N1 = 194, N2 = 43. Standard errors are in parentheses. ^***p < .001 ^** p < .01 ^* p < .05 ^† p < .10

在文檔中 以階層線性模式分析個人與組織因素對學習行為之效果：以組織因素為調節角色 (頁 64-72)