- Email: [email protected]
Configural theory for ICT development
Journal of Business Research 68 (2015) 748–756
Contents lists available at ScienceDirect
Journal of Business Research
Configural theory for ICT development☆ Kun-Huang Huarng Department of International Trade, Feng Chia University, 100 Wenhwa Road, Seatwen, Taichung 40724, Taiwan
a r t i c l e
i n f o
Available online 9 December 2014 Keywords: Causal complexity Corruption Fuzzy set/Qualitative Comparative Analysis (fsQCA) Population density Predictive validity
a b s t r a c t This study intends to establish configural theory for ICT development by using fuzzy set/Qualitative Comparative Analysis (fsQCA) and to contrast the results with those from multivariate regression analysis (MRA). The fsQCA results support three propositions: the highly-developed countries, the highly-developed countries with low population density and the highly-developed countries with low corruption are the sufficient conditions for high ICT development. In addition, the improvement toward developed countries and increases in both population density and corruption are a sufficient condition for the improvement in ICT development. However, fsQCA finds a contrary case: the improvement toward developed countries and decreases in both population density and corruption are also a sufficient condition for the improvement in ICT development. MRA is good at model fitting. FsQCA is good at showing the causal complexities to explain the outcome and successfully predicts the withheld data sets. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Information and communications technology (ICT) impacts countries all over the world (Huarng, 2010; Huarng & Yu, 2011). Some studies examine ICT adoption at the country level, while others on firm level (Kim & Huarng, 2011). Yu (2011) points out that the least squares method only considers the impact of the estimated coefficients on the means and she presents a quantile model to examine the impacts of each variable at different quantiles on ICT development. Hence, this study aims to establish configural theory for ICT development and to contrast the analysis with multiple regression analysis (MRA). Woodside, Camacho, and Lai (2013) state that the evaluation of outcomes requires the development of the configuration of causes and processes that lead to the outcomes. Hence, the use of appropriate methods is important in identifying correct relationships. The conventional statistical methods tend to report the “net effects,” that is the direct plus indirect influence of each independent variable on the dependent variable (Woodside, Schpektor, & Xia, 2013). These methods are suitable for symmetric data, such as in Fig. 1A, where there are high values of X (independent variable) for high values of Y (dependent variable), such as a and b, and low values of X for low values of Y, such as c and d. However, Ragin (2008) expresses two concerns. First, the combination of three to six independent variables presents a level of complexity that the
☆ The author acknowledges and is grateful for the financial support provided by the National Science Council, Taiwan, ROC under grant NSC 102-2410-H-035-038-MY2. The authors would also like to thank Arch Woodside, Boston College, USA and Editor of the Journal of Business Research for his valuable suggestions on how to construct the fsQCA empirical models and analyses. E-mail address: [email protected].
http://dx.doi.org/10.1016/j.jbusres.2014.11.023 0148-2963/© 2014 Elsevier Inc. All rights reserved.
statistical modeling of three- to six-way interactions in MRA cannot easily implement. Second, the real-life relationships of independent variables and dependent variables are asymmetric. Fig. 1B shows the asymmetric data relationships in that on top of the situation as in Fig. 1A, there are also low values of X for high values of Y, such as e, f, and g. Hence, this study uses fuzzy set/Qualitative Comparative Analysis (fsQCA) to explore the relationships between the antecedents (the dependent variables) and the outcome (the independent variable). In addition, this study also explores the relationships between the improvements in the antecedents and the improvement in ICT development. FsQCA suits the problem with asymmetric data and identifies the relationships between different combinations of antecedents and the outcome. To compare fsQCA with MRA, this study also conducts similar analyses. This study then compares the results and discusses the different capabilities of and results of these two methods. Section 2 proposes the relevant propositions. Section 3 introduces the methods, as well as the variables and data. Section 4 reports and compares the empirical analyses by using both fsQCA and MRA. This study also explores the meaning of each combination of antecedents. Section 5 concludes this article. 2. Propositions Different factors may impact ICT development. Various studies associate developed countries with high ICT development (Kasikitsakunphon & Vanijja, 2013; Raghuprasad, Devaraja, & Gopala, 2013). The economic level and the level of industrialization are the two fundamental factors to determine whether a country is developed (Investopedia, 2014). Many studies use GDP per capita (Huarng, 2011; Huarng & Yu, 2014)
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
749
P3. Highly-developed countries with low corruption → high ICT development.
A)
This study considers that similar antecedent combinations would apply to the improvement in ICT development. Following the above propositions, this study also puts forward a proposition for the improvement in ICT development: P4. Improvement toward developed countries and decrease in population density and decrease in corruption → improvement in ICT development. 3. Research methods 3.1. Methods This study compares the empirical results from both MRA and fsQCA. The MRA includes correlation analysis, regression analysis, and the analysis of variance. FsQCA is an analytical tool that uses fuzzy set theory and Boolean logic. FsQCA is different from the conventional statistical methods in the following ways: set-theoretic vs. correlational connections, calibration vs. measurement, configuration of conditions vs. independent variables, and analysis of causal complexity vs. the analysis of net effects (Ragin, 2008). FsQCA first calibrates the data into 0.0–1.0, where 0.0 means full non-membership and 1.0 means full membership. In considering the thresholds, this study takes b0.05 as full nonmembership and N0.95 as full membership. Both Ragin (2008) and Woodside and Zhang (2013) provide more detail on how to perform the calibrations. After processing the calibrated data, fsQCA provides the configuration of conditions, such as
B)
A B→C
Fig. 1. A. Symmetric data. B. Asymmetric data.
or GDP per capita in purchasing power parity (PPP) terms (Balaban, 2012; Mehmood & Azim, 2013) to represent a country's economic level and explore the relationships with ICT development. Studies also use high numbers of telephone lines per 100 inhabitants to represent industrialization (Hudson, 1995; Hudson, 2000). Hence, this study proposes using PPP to represent a country's economic level and telephone lines per 100 inhabitants to represent the level of industrialization and puts forward the following propositions: High PPP and high telephone lines per 100 inhabitants → highlydeveloped countries. P1. Highly-developed countries → high ICT development where → means sufficient conditions. Two other variables are also of interest. What is the relationship between population density and ICT development? Some studies suggest that low population density contributes to high ICT development (DeMaagd, 2009). The people in countries with low population densities may desire more communication from ICT. Hence, this study proposes P2. Highly-developed countries with low population density → high ICT development. Most studies focus on how ICT can reduce corruption (Charoensukmongkol & Moqbel, 2014; Ionescu, 2013). Hence, this study intends to explore how corruption may affect ICT development in different countries and proposes.
where A and B are the antecedents, and C is the outcome. * represents logic AND, while ~ represents NOT. The equation means that A* ~ B (an antecedent combination) is the sufficient condition for C. Both Ragin (2008) and Woodside (2013) stress the importance of achieving high consistency over the high coverage. Hence, this study focuses more on the consistency although also encompasses the coverage. 3.2. Variables and data This study explains the variables in Table 1, including the outcome, namely, the Internet, and the antecedents, i.e., ppp, phone, pop, and cpi. To study the improvement in antecedents and outcome, this study adds a d in front of each antecedent to represent the difference; for example, d_ppp is the difference in the ppp of two time periods. This study also adds c to each antecedent to represent the calibrated antecedent; for example, c_ppp is the calibrated ppp. The data for the Internet, phone, and pop are from the data set of the World Telecommunication/ICT Indicators database compiled by the
Table 1 Variable names for outcome and antecedents. Outcome/variable Definition name Internet
Note
Internet users per 100 inhabitants High value indicates to represent ICT development high ICT development. Antecedents/variable names ppp GDP per capita in purchasing power parity phone Telephone lines per 100 inhabitants pop Population density cpi Corruption index High value indicates low corruption
750
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
International Telecommunication Union (ITU, 2002). The data for ppp are from the World Bank (World Bank, 2014), while the data for the cpi are from Transparency International (Transparency International, 2014). The analysis ranges over the period from 2004 to 2011. There are more than 110 data entries for each type of analysis. 4. Empirical analyses This study conducts three sets of empirical analysis, including the 3year lag analysis, prediction, and the improvement analysis using both fsQCA and MRA. This study also compares the capabilities of both fsQCA and MRA.
The remaining tables in Table 2 show the results for various year patterns. The consistency of each individual antecedent combination is greater than 0.950. The solution consistency of each year pattern is also greater than 0.950. This study points out only the repetitive antecedent combinations in different year patterns, but in Table 2 there is an antecedent combination, c_ppp*c_phone in the year patterns of 2006–2009 and 2007– 1010. In other words, c ppp c phone → Internet which supports the proposition:
4.1. Three-year lag analysis
P1. Highly-developed countries → high ICT development.
First, this study uses a 3-year lag analysis to show how the antecedents in the time period t − 3 impact the outcome in t. The equation for the analysis is
In the year patterns of 2004–2007, 2005–2008, and 2008–2011, there is an antecedent combination, c_ppp* ~ c_pop*c_phone, which supports proposition
Internetðt Þ ¼ f ðphoneðt−3Þ; popðt−3Þ; pppðt−3Þ; cpiðt−3ÞÞ:
P2. Highly-developed countries and low population density → high ICT development.
For example, 2004–2007 represents a year pattern for Internetð2007Þ ¼ f ðphoneð2004Þ; popð2004Þ; pppð2004Þ; cpið2004ÞÞ: Table 2 lists the empirical results from fsQCA for different year patterns. The first table in Table 2 is for the year pattern 2004–2007, or Internet(2007) = f(phone(2004), pop(2004), ppp(2004), cpi(2004)). FsQCA suggests the logical OR of two antecedent combinations from the analysis: c ppp c pop c phone OR c cpi c ppp c phone whose consistencies are 0.975 and 0.980, respectively. Both consistencies are greater than 0.970. The solution coverage and solution consistency of c_ppp* ~ c_pop*c_phone OR c_cpi*c_ppp*c_phone are 0.843 and 0.974, respectively. Table 2 Three-year lag results from fsQCA.
In the year patterns of 2004–2007 and 2005–2008, there is an antecedent combination, c_cpi*c_ppp*c_phone, which supports proposition P3. Highly-developed countries and low corruption → high ICT development. Table 2 also presents other antecedent combinations, which show the causal complexities of ICT development. Table 3 lists the analyses using MRA. All the adjusted R2 values are greater than 0.880, which shows that MRA has good model fitting capability. All terms have positive impacts on the Internet; however, not all terms are significant in each year pattern. 4.2. Predictive validities To demonstrate the predictive validities using fsQCA, this study performs the prediction using a 3-year lag analysis. This study predicts the rest of the year patterns by using the year pattern for 2004–2007: c ppp c pop c phone OR c cpi c ppp c phone
Raw coverage
Unique coverage
Consistency
Internet(2007) = f(phone(2004), pop(2004), ppp(2004), cpi(2004)) c_ppp* ~ c_pop*c_phone 0.601 0.055 0.975 c_cpi*c_ppp*c_phone 0.788 0.242 0.980 Solution coverage: 0.843 Solution consistency: 0.974 Internet(2008) = f(phone(2005), pop(2005), ppp(2005), cpi(2005)) c_ppp* ~ c_pop*c_phone 0.613 0.050 0.967 c_cpi*c_ppp*c_phone 0.812 0.248 0.972 Solution coverage: 0.861 Solution consistency: 0.966 Internet(2009) = f(phone(2006), pop(2006), ppp(2006), cpi(2006)) c_ppp*c_phone 0.862 0.862 0.972 Solution coverage: 0.862 Solution consistency: 0.972 Internet(2010) = f(phone(2007), pop(2007), ppp(2007), cpi(2007)) c_ppp*c_phone 0.854 0.323 0.977 c_cpi*c_pop*c_phone 0.541 0.010 0.964 Solution coverage: 0.864 Solution consistency: 0.965 Internet(2011) = f(phone(2008), pop(2008), ppp(2008), cpi(2008)) c_cpi* ~ c_pop*c_phone 0.580 0.020 c_ppp* ~ c_pop*c_phone 0.619 0.039 c_cpi*c_ppp*c_pop 0.525 0.107 ~c_cpi* ~ c_ppp*c_pop*c_phone 0.320 0.022 Solution coverage: 0.789 Solution consistency: 0.956
0.978 0.978 0.972 0.952
Table 4 lists the results of the prediction. The consistency of each antecedent combination is greater than 0.960. The solution consistency of each year pattern is also greater than 0.960 and the solution coverage is greater than 0.820. All of these findings show that fsQCA possesses great predicting capabilities. Similarly, this study also uses the equation from the year pattern 2004–2007 to predict the rest of the year patterns by using MRA: Internet = −6.038 + 0.322*phone + 0.002*pop + 0.000*ppp + 1.680*cpi. Table 5 shows the predictive validities using MRA. None of the adjusted R2 values are greater than 0.015 and none of the items in each year is of significance, which shows that MRA is poor in terms of prediction. 4.3. Improvement in ICT development To understand the impacts of antecedent combinations on the improvement in ICT development, this study uses both fsQCA and MRA to perform analyses for the differences in the 3-year lag. The first analysis is
Internetð2010Þ−Internetð2007Þ ¼ fðphoneð2007Þ–phoneð2004Þ; popð2007Þ–popð2004Þ; pppð2007Þ–pppð2004Þ; cpið2007Þ–cpið2004ÞÞ:
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
751
Table 3 Three-year lag results from MRA. Model summary (2004–2007) Model
R
1 .949a a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.901
.897
3.879
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
14231.000 1564.770 15795.000
3557.663 15.046
236.450
b0.000 b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −6.038 phone 1 0.322 pop 1 0.002 ppp 1 0.000 cpi 1 1.680 a. Dependent variable: Internet Internet = −6.038 + 0.322*phone + 0.002*pop + 0.000*ppp + 1.680*cpi
SE
t-Value
Sig Pr N |t|
VIF
0.983 0.039 0.000 0.000 0.350
−6.140 8.300 4.560 1.630 4.800
b0.000* b0.000* b0.000* 0.106 b0.000*
0.000 4.321 1.092 4.069 4.366
Model summary (2005–2008) Model
R a
1 .941 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.886
.881
4.038
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
13121.000 1695.999 14817.000
3280.341 16.308
201.150
b0.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −4.549 phone 1 0.388 pop 1 0.000 ppp 1 0.000 cpi 1 1.146 a. Dependent variable: Internet Internet = −4.549 + 0.388*phone + 0.000*pop + 0.000*ppp + 1.146*cpi
SE
t-Value
Sig Pr N |t|
VIF
0.986 0.040 0.000 0.000 0.352
−4.610 9.630 0.260 1.930 3.260
b0.000* b0.000* 0.796 0.056 0.002*
0.000 3.923 1.102 4.168 4.152
Model summary (2006–2009) Model
R a
1 .944 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.892
.888
4.002
ANOVAa Model
DF
Regression 4 Residual 109 Total 113 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
14405.000 1745.467 16150.000
3601.131 16.013
224.880
b0.000b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop
1 1 1
−3.753 0.391 0.000
0.961 0.039 0.000
−3.91 10.03 0.57
0.000* b0.000* 0.567
0.000 3.932 1.100 (continued on next page)
752
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
Table 3 (continued) Coefficientsa Independent variables
DF
Coefficient
ppp 1 0.000 cpi 1 0.895 a. Dependent variable: Internet Internet = −3.753 + 0.391*phone + 0.000*pop + 0.000*ppp + 0.895*cpi
SE
t-Value
0.000 0.337
Sig Pr N |t|
2.81 2.66
VIF
0.006* 0.009*
4.582 4.068
Model summary (2007–2010) Model
R a
1 .956 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.914
.911
3.689
ANOVAa Model
DF
Regression 4 Residual 107 Total 111 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
15517.000 1456.313 16973.000
3879.286 13.610
285.020
b.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −4.416 phone 1 0.485 pop 1 0.000 ppp 1 0.000 cpi 1 0.997 a. Dependent variable: Internet Internet = −4.416 + 0.485*phone + 0.000*pop + 0.000*ppp + 0.997*cpi
SE
t-Value
Sig Pr N |t|
VIF
0.918 0.035 0.000 0.000 0.320
−4.81 14.06 0.70 1.46 3.12
b0.000* b0.000* 0.483 0.147 0.002*
0.000 3.408 1.085 3.568 4.059
Model summary (2008–2011) Model
R a
1 .948 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.899
.895
3.923
ANOVAa Model
DF
Regression 4 Residual 93 Total 97 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
12789.000 1431.466 14220.000
3197.128 15.392
207.710
b.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −4.292 phone 1 0.534 pop 1 0.000 ppp 1 0.000 cpi 1 0.956 a. Dependent variable: Internet Internet = −4.292 + 0.534*phone + 0.000*pop + 0.000*ppp + 0.956*cpi
The second analysis is Internetð2011Þ‐Internetð2008Þ ¼ fðphoneð2008Þ–phoneð2005Þ; popð2008Þ–popð2005Þ; pppð2008Þ–pppð2005Þ; cpið2008Þ–cpið2005ÞÞ:
In the first table of Table 6, fsQCA suggests the following three antecedent combinations: c_d_cpi*c_d_ppp*c_d_pop* ~ c_d_phone, c_d_cpi*c_d_ppp* ~ c_d_pop*c_d_phone, OR ~
SE
t-Value
Sig Pr N |t|
VIF
0.958 0.035 0.000 0.000 0.285
−4.480 15.450 0.160 0.700 3.360
b0.000* b0.000* 0.871 0.484 0.001*
0.000 0.412 0.935 0.372 0.417
c_d_cpi*c_d_ppp*c_d_pop*c_d_phone with consistency of 0.916, 0.919, and 0.926, respectively. The solution consistency is 0.904. There is only one antecedent combination in the second analysis, ~c_d_cpi*c_d_ppp*c_d_pop*c_d_phone. When this study examines the recurrent antecedent combinations, there is only one antecedent combination in both analyses: c d cpi c d ppp c d pop c d phone which does not support P4.
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
753
However, there is an antecedent combination in the first analysis:
Table 4 Predictive validities by fsQCA. Raw coverage
Unique coverage
Consistency
Internet(2008) = f(phone(2005), pop(2005), ppp(2005), cpi(2005)) c_ppp* ~ c_pop*c_phone 0.613 0.050 0.967 c_cpi*c_ppp*c_phone 0.812 0.248 0.972 Solution coverage: 0.861 Solution consistency: 0.966 Internet(2009) = f(phone(2006), pop(2006), ppp(2006), cpi(2006)) c_ppp* ~ c_pop*c_phone 0.618 0.047 0.974 c_cpi*c_ppp*c_phone 0.798 0.226 0.983 Solution coverage: 0.845 Solution consistency: 0.977 Internet(2010) = f(phone(2007), pop(2007), ppp(2007), cpi(2007)) c_ppp* ~ c_pop*c_phone 0.620 0.040 0.980 c_cpi*c_ppp*c_phone 0.799 0.219 0.981 Solution coverage: 0.839 Solution consistency: 0.978 Internet(2011) = f(phone(2008), pop(2008), ppp(2008), cpi(2008)) c_ppp* ~ c_pop*c_phone 0.628 0.067 0.978 c_cpi*c_ppp*c_phone 0.760 0.200 0.986 Solution coverage: 0.827 Solution consistency: 0.980
c d cpi c d ppp c d pop c d phone
which supports P4. These two antecedent combinations are rather different. The former one increases in both corruption and population density, while the latter one decreases in both corruption and population density. However, contrary cases exist in the same year with high consistencies. This demonstrates that fsQCA can present contrary cases at the same time to explain the same outcome. The third antecedent combination in the first analysis is c_d_cpi*c_d_ppp*c_d_pop* ~ c_d_phone, which represents an improvement toward a rich country, an increase in population density, but a decrease in corruption and phone lines. Table 7 shows the analyses of differences using MRA. As in both tables in Table 7, both d_phone and d_ppp offer positive impacts but d_ppp and d_cpi negative impacts on the improvement in the Internet. However, the only significant variable is pop in the first analysis and ppp in the second analysis, respectively.
Table 5 Predictive validities by MRA. Model summary (2008) Model
R a
1 .205 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.042
.005
11.131
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
563.660 12885.000 13449.000
140.915 123.893
1.140
0.343b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop ppp cpi a. Dependent variable: Internet
1 1 1 1 1
4.696 −0.044 −0.001 −0.000 1.809
2.820 0.111 0.001 0.000 1.006
1.670 −0.390 −0.810 −0.700 1.800
0.099 0.696 0.420 0.486 0.075
0.000 4.322 1.093 4.070 4.366
Model summary (2009) Model
R a
1 .071 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.005
−0.033
12.084
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
77.762 15186.000 15264.000
19.440 146.018
0.130
0.970b
Coefficientsa Independent variables
DF
Intercept phone
1 1
Coefficient 12.793 0.045
SE 3.062 0.121
t-Value
Sig Pr N |t|
VIF
4.180 0.370
b0.000 0.713
0.000 4.322 (continued on next page)
754
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
Table 5 (continued) Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
pop ppp cpi a. Dependent variable: Internet
1 1 1
0.000 −0.000 −0.611
0.001 0.000 1.092
0.340 −0.060 −0.560
0.733 0.951 0.577
VIF 1.093 4.070 4.367
Model summary (2010) Model
R a
1 .152 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.023
−0.017
12.090
ANOVAa Model
DF
Regression 4 Residual 98 Total 102 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
337.406 14325.000 14662.000
84.351 146.170
0.580
0.680b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop ppp cpi a. Dependent variable: Internet
1 1 1 1 1
14.652 0.131 0.000 −0.000 −1.274
3.171 0.122 0.001 0.000 1.117
4.620 1.070 0.290 −0.350 −1.140
b0.000 0.285 0.774 0.724 0.257
0.000 4.223 1.092 4.092 4.370
Model summary (2011) Model
R a
1 .244 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.059
0.015
12.037
ANOVAa Model
DF
Regression 4 Residual 84 Total 88 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
767.162 12170.000 12937.000
191.791 144.881
1.320
0.268b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop ppp cpi a. Dependent variable: Internet
1 1 1 1 1
12.591 −0.095 −0.002 0.000 −0.869
3.549 0.142 0.001 0.000 1.277
3.550 −0.670 −1.050 1.800 −0.680
0.001 0.504 0.297 0.076 0.498
0.000 4.654 1.139 6.830 4.764
5. Discussion and conclusion This study establishes configural theory for ICT development using fsQCA and compares the capabilities of fsQCA and MRA. The fsQCA results support three propositions: the highly-developed countries, the highly-developed countries with low population density, and the highly-developed countries with low corruption are the sufficient conditions for high ICT development. Fig. 2 depicts the possible relationships. FsQCA shows the causal complexities for the same outcome, while MRA is good at model fitting but falls short in terms of presenting the causal complexities.
This study separates the data and uses the first year pattern to obtain the relationships and then to predict the patterns for the rest of the years. FsQCA shows very strong capabilities in prediction. On the contrary, MRA predicts poorly. As for the improvement in ICT development, fsQCA finds that the improvement toward developed countries and increases in both population density and corruption are the sufficient condition for the improvement in ICT development for both analyses which do not support the proposition. Interestingly, fsQCA also finds a contrary case which is also a sufficient condition for the improvement in ICT development: the improvement toward developed countries and the decrease
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
755
Table 6 Improvement analyses fsQCA. Raw coverage
Unique coverage
Consistency
Internet(2010) − Internet(2007) = f(phone(2007) – phone(2004), pop(2007) – pop(2004), ppp(2007) – ppp(2004), cpi(2007) – cpi(2004)) c_d_cpi*c_d_ppp*c_d_pop* ~ c_d_phone 0.264 0.047 0.916 c_d_cpi*c_d_ppp* ~ c_d_pop*c_d_phone 0.322 0.093 0.919 ~c_d_cpi*c_d_ppp*c_d_pop*c_d_phone 0.255 0.058 0.926 Solution coverage: 0.429 Solution consistency: 0.904 Internet(2011) − Internet(2008) = f(phone(2008) – phone(2005), pop(2008) – pop(2005), ppp(2008) – ppp(2005), cpi(2008) – cpi(2005)) ~c_d_cpi*c_d_ppp*c_d_pop*c_d_phone 0.298 0.298 0.911 Solution coverage: 0.298 Solution consistency: 0.911
Fig. 2. Possible relationships between the antecedent combinations and the outcome.
Table 7 Improvement analyses by MRA. Model summary Internet(2010) − Internet(2007) = f(phone(2007) – phone(2004), pop(2007) – pop(2004), ppp(2007) – ppp(2004), cpi(2007) – cpi(2004)) Model
R2
R
Adjusted R2
Std. error of the estimate
1 .645a 0.416 0.382 2.687 a. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
References
ANOVAa Model
DF
Sum of squares
Mean square
Regression 4 354.251 88.563 Residual 69 497.940 7.217 Total 73 852.191 a. Dependent variable: d_Internet b. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
F
Sig (Pr N F)
12.272
b.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 2.233 phone 1 0.088 pop 1 −0.031 ppp 1 0.000 cpi 1 0.289 a. Dependent variable: Internet
SE
t-Value
Sig Pr N |t|
VIF
0.434 0.085 0.005 0.000 0.688
5.140 1.040 −6.550 0.680 0.420
b0.000 0.301 b0.000* 0.499 0.676
0.000 1.041 1.123 1.114 1.034
Model summary Internet(2011) − Internet(2008) = f(phone(2008) – phone(2005), pop(2008) – pop(2005), ppp(2008) – ppp(2005), cpi(2008) – cpi(2005)) Model
R2
R
Adjusted R2
Std. error of the estimate
a
0.344 0.308 1 .587 a. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
4.464
ANOVAa Model
in both population density and corruption. This antecedent combination supports Proposition 4. FsQCA can show the contrary cases to explain the same outcome while MRA is incapable of doing so.
DF
Sum of squares
Mean square
Regression 4 753.269 188.317 Residual 72 1434.799 19.928 Total 76 2188.068 a. Dependent variable: d_Internet b. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
F
Sig (Pr N F)
9.450
b0.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 3.382 phone 1 0.071 pop 1 −0.035 ppp 1 0.000 cpi 1 −0.727 a. Dependent variable: Internet
SE
t-Value
Sig Pr N |t|
VIF
0.660 0.155 0.006 0.000 0.695
5.130 0.460 −5.890 1.450 −1.050
b0.000 0.647 b0.000* 0.153 0.299
0.000 1.027 1.043 1.035 1.023
Balaban, M. E. (2012). Evaluation of human development index and ICT development index, comparative analysis of the OECD and the European members and Turkey. International Journal of Business Research, 12(6), 126–130. Charoensukmongkol, P., & Moqbel, M. (2014). Does investment in ICT curb or create more corruption? Across-country analysis. Public Organization Review, 14(1), 51–63. DeMaagd, K. (2009). The myth of population density and ICT infrastructure. HICSS '09. 42nd Hawaii International Conference on System Sciences (pp. 1–10). Huarng, K. -H. (2010). Essential research in technology management. Journal of Business Research, 63(5), 451–453. Huarng, K. -H. (2011). A comparative study to classify ICT developments by economies. Journal of Business Research, 64(11), 1174–1177. Huarng, K. -H., & Yu, T. H. -K. (2011). Internet software and services: Past and future. Service Industries Journal, 31(1–2), 79–89. Huarng, K. -H., & Yu, T. H. -K. (2014). A new quantile regression forecasting model. Journal of Business Research, 67(5), 779–784. Hudson, H. E. (1995). Access to telecommunications in the developing world: Ten years after the Maitland report. In G. W. Brock (Ed.), Toward a Competitive Telecommunication Industry: Selected Papers From the 1994 Telecommunications Policy Research Conference (LEA Telecommunications Series) (pp. 235–248). Routledge. Hudson, H. E. (2000). Global information infrastructure: Eliminating the distance barrier. In M. Lucas (Ed.), Understanding business: Environments (pp. 235–244). International Telecommunication Union (ITU) (2002). World telecommunications development report. Investopedia (2014). Developed economy. http://www.investopedia.com/terms/d/ developed-economy.asp Ionescu, L. (2013). The positive effect of ICT infrastructure in reducing corruption and increasing transparency. Economics, Management, and Financial Markets, 2, 167–172. Kasikitsakunphon, S., & Vanijja, V. (2013). Factors influencing the intention to use VoIP service by consumers in Thailand. 2013 International Conference on ICT Convergence (ICTC) (pp. 997–1002). Kim, S. -H., & Huarng, K. -H. (2011). Winning strategies for innovation and hightechnology products management. Journal of Business Research, 64(11), 1147–1150. Mehmood, B., & Azim, P. (2013). Does ICT participate in economic convergence among Asian countries: Evidence from dynamic panel data model. Informatica Economică, 17(2), 7–16. Raghuprasad, K. P., Devaraja, S. C., & Gopala, Y. M. (2013). An analysis of knowledge level of farmers on utilization of ICT tools for farm communication. Journal of Rural Development, 32(3), 301–310. Ragin, C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago: Chicago University Press. Transparency International (2014). Corruption Perceptions Index, Overview. http://www. transparency.org/research/cpi/overview Woodside, A. G. (2013). Moving beyond multiple regression analysis to algorithms: Calling for a paradigm shift from symmetric to asymmetric thinking in data analysis and crafting theory. Journal of Business Research, 63, 463–472. Woodside, A. G., Camacho, A. R., & Lai, W. -H. (2013). Sense making, dilemma, and solutions in strategic management. International Journal of Business and Economics, 12(2), 91–95.
756
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
Woodside, A. G., Schpektor, A., & Xia, X. (2013). Triple sense-making of findings from marketing experiments using the dominant variable-based logic, case-based logic, and isomorphic modeling. International Journal of Business and Economics, 12(2), 131–154. Woodside, A. G., & Zhang, M. (2013). Cultural diversity and marketing transactions: Are market integration, large community size, and world religions necessary for fairness in ephemeral exchanges? Psychology & Marketing, 30(3), 263–276.
World Bank (2014). GNI per capita, PPP (current international $). http://data.worldbank. org/indicator/NY.GNP.PCAP.PP.CD Yu, T. H. -K. (2011). Heterogeneous effects of different factors on global ICT adoption. Journal of Business Research, 64(11), 1169–1173.
Contents lists available at ScienceDirect
Journal of Business Research
Configural theory for ICT development☆ Kun-Huang Huarng Department of International Trade, Feng Chia University, 100 Wenhwa Road, Seatwen, Taichung 40724, Taiwan
a r t i c l e
i n f o
Available online 9 December 2014 Keywords: Causal complexity Corruption Fuzzy set/Qualitative Comparative Analysis (fsQCA) Population density Predictive validity
a b s t r a c t This study intends to establish configural theory for ICT development by using fuzzy set/Qualitative Comparative Analysis (fsQCA) and to contrast the results with those from multivariate regression analysis (MRA). The fsQCA results support three propositions: the highly-developed countries, the highly-developed countries with low population density and the highly-developed countries with low corruption are the sufficient conditions for high ICT development. In addition, the improvement toward developed countries and increases in both population density and corruption are a sufficient condition for the improvement in ICT development. However, fsQCA finds a contrary case: the improvement toward developed countries and decreases in both population density and corruption are also a sufficient condition for the improvement in ICT development. MRA is good at model fitting. FsQCA is good at showing the causal complexities to explain the outcome and successfully predicts the withheld data sets. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Information and communications technology (ICT) impacts countries all over the world (Huarng, 2010; Huarng & Yu, 2011). Some studies examine ICT adoption at the country level, while others on firm level (Kim & Huarng, 2011). Yu (2011) points out that the least squares method only considers the impact of the estimated coefficients on the means and she presents a quantile model to examine the impacts of each variable at different quantiles on ICT development. Hence, this study aims to establish configural theory for ICT development and to contrast the analysis with multiple regression analysis (MRA). Woodside, Camacho, and Lai (2013) state that the evaluation of outcomes requires the development of the configuration of causes and processes that lead to the outcomes. Hence, the use of appropriate methods is important in identifying correct relationships. The conventional statistical methods tend to report the “net effects,” that is the direct plus indirect influence of each independent variable on the dependent variable (Woodside, Schpektor, & Xia, 2013). These methods are suitable for symmetric data, such as in Fig. 1A, where there are high values of X (independent variable) for high values of Y (dependent variable), such as a and b, and low values of X for low values of Y, such as c and d. However, Ragin (2008) expresses two concerns. First, the combination of three to six independent variables presents a level of complexity that the
☆ The author acknowledges and is grateful for the financial support provided by the National Science Council, Taiwan, ROC under grant NSC 102-2410-H-035-038-MY2. The authors would also like to thank Arch Woodside, Boston College, USA and Editor of the Journal of Business Research for his valuable suggestions on how to construct the fsQCA empirical models and analyses. E-mail address: [email protected].
http://dx.doi.org/10.1016/j.jbusres.2014.11.023 0148-2963/© 2014 Elsevier Inc. All rights reserved.
statistical modeling of three- to six-way interactions in MRA cannot easily implement. Second, the real-life relationships of independent variables and dependent variables are asymmetric. Fig. 1B shows the asymmetric data relationships in that on top of the situation as in Fig. 1A, there are also low values of X for high values of Y, such as e, f, and g. Hence, this study uses fuzzy set/Qualitative Comparative Analysis (fsQCA) to explore the relationships between the antecedents (the dependent variables) and the outcome (the independent variable). In addition, this study also explores the relationships between the improvements in the antecedents and the improvement in ICT development. FsQCA suits the problem with asymmetric data and identifies the relationships between different combinations of antecedents and the outcome. To compare fsQCA with MRA, this study also conducts similar analyses. This study then compares the results and discusses the different capabilities of and results of these two methods. Section 2 proposes the relevant propositions. Section 3 introduces the methods, as well as the variables and data. Section 4 reports and compares the empirical analyses by using both fsQCA and MRA. This study also explores the meaning of each combination of antecedents. Section 5 concludes this article. 2. Propositions Different factors may impact ICT development. Various studies associate developed countries with high ICT development (Kasikitsakunphon & Vanijja, 2013; Raghuprasad, Devaraja, & Gopala, 2013). The economic level and the level of industrialization are the two fundamental factors to determine whether a country is developed (Investopedia, 2014). Many studies use GDP per capita (Huarng, 2011; Huarng & Yu, 2014)
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
749
P3. Highly-developed countries with low corruption → high ICT development.
A)
This study considers that similar antecedent combinations would apply to the improvement in ICT development. Following the above propositions, this study also puts forward a proposition for the improvement in ICT development: P4. Improvement toward developed countries and decrease in population density and decrease in corruption → improvement in ICT development. 3. Research methods 3.1. Methods This study compares the empirical results from both MRA and fsQCA. The MRA includes correlation analysis, regression analysis, and the analysis of variance. FsQCA is an analytical tool that uses fuzzy set theory and Boolean logic. FsQCA is different from the conventional statistical methods in the following ways: set-theoretic vs. correlational connections, calibration vs. measurement, configuration of conditions vs. independent variables, and analysis of causal complexity vs. the analysis of net effects (Ragin, 2008). FsQCA first calibrates the data into 0.0–1.0, where 0.0 means full non-membership and 1.0 means full membership. In considering the thresholds, this study takes b0.05 as full nonmembership and N0.95 as full membership. Both Ragin (2008) and Woodside and Zhang (2013) provide more detail on how to perform the calibrations. After processing the calibrated data, fsQCA provides the configuration of conditions, such as
B)
A B→C
Fig. 1. A. Symmetric data. B. Asymmetric data.
or GDP per capita in purchasing power parity (PPP) terms (Balaban, 2012; Mehmood & Azim, 2013) to represent a country's economic level and explore the relationships with ICT development. Studies also use high numbers of telephone lines per 100 inhabitants to represent industrialization (Hudson, 1995; Hudson, 2000). Hence, this study proposes using PPP to represent a country's economic level and telephone lines per 100 inhabitants to represent the level of industrialization and puts forward the following propositions: High PPP and high telephone lines per 100 inhabitants → highlydeveloped countries. P1. Highly-developed countries → high ICT development where → means sufficient conditions. Two other variables are also of interest. What is the relationship between population density and ICT development? Some studies suggest that low population density contributes to high ICT development (DeMaagd, 2009). The people in countries with low population densities may desire more communication from ICT. Hence, this study proposes P2. Highly-developed countries with low population density → high ICT development. Most studies focus on how ICT can reduce corruption (Charoensukmongkol & Moqbel, 2014; Ionescu, 2013). Hence, this study intends to explore how corruption may affect ICT development in different countries and proposes.
where A and B are the antecedents, and C is the outcome. * represents logic AND, while ~ represents NOT. The equation means that A* ~ B (an antecedent combination) is the sufficient condition for C. Both Ragin (2008) and Woodside (2013) stress the importance of achieving high consistency over the high coverage. Hence, this study focuses more on the consistency although also encompasses the coverage. 3.2. Variables and data This study explains the variables in Table 1, including the outcome, namely, the Internet, and the antecedents, i.e., ppp, phone, pop, and cpi. To study the improvement in antecedents and outcome, this study adds a d in front of each antecedent to represent the difference; for example, d_ppp is the difference in the ppp of two time periods. This study also adds c to each antecedent to represent the calibrated antecedent; for example, c_ppp is the calibrated ppp. The data for the Internet, phone, and pop are from the data set of the World Telecommunication/ICT Indicators database compiled by the
Table 1 Variable names for outcome and antecedents. Outcome/variable Definition name Internet
Note
Internet users per 100 inhabitants High value indicates to represent ICT development high ICT development. Antecedents/variable names ppp GDP per capita in purchasing power parity phone Telephone lines per 100 inhabitants pop Population density cpi Corruption index High value indicates low corruption
750
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
International Telecommunication Union (ITU, 2002). The data for ppp are from the World Bank (World Bank, 2014), while the data for the cpi are from Transparency International (Transparency International, 2014). The analysis ranges over the period from 2004 to 2011. There are more than 110 data entries for each type of analysis. 4. Empirical analyses This study conducts three sets of empirical analysis, including the 3year lag analysis, prediction, and the improvement analysis using both fsQCA and MRA. This study also compares the capabilities of both fsQCA and MRA.
The remaining tables in Table 2 show the results for various year patterns. The consistency of each individual antecedent combination is greater than 0.950. The solution consistency of each year pattern is also greater than 0.950. This study points out only the repetitive antecedent combinations in different year patterns, but in Table 2 there is an antecedent combination, c_ppp*c_phone in the year patterns of 2006–2009 and 2007– 1010. In other words, c ppp c phone → Internet which supports the proposition:
4.1. Three-year lag analysis
P1. Highly-developed countries → high ICT development.
First, this study uses a 3-year lag analysis to show how the antecedents in the time period t − 3 impact the outcome in t. The equation for the analysis is
In the year patterns of 2004–2007, 2005–2008, and 2008–2011, there is an antecedent combination, c_ppp* ~ c_pop*c_phone, which supports proposition
Internetðt Þ ¼ f ðphoneðt−3Þ; popðt−3Þ; pppðt−3Þ; cpiðt−3ÞÞ:
P2. Highly-developed countries and low population density → high ICT development.
For example, 2004–2007 represents a year pattern for Internetð2007Þ ¼ f ðphoneð2004Þ; popð2004Þ; pppð2004Þ; cpið2004ÞÞ: Table 2 lists the empirical results from fsQCA for different year patterns. The first table in Table 2 is for the year pattern 2004–2007, or Internet(2007) = f(phone(2004), pop(2004), ppp(2004), cpi(2004)). FsQCA suggests the logical OR of two antecedent combinations from the analysis: c ppp c pop c phone OR c cpi c ppp c phone whose consistencies are 0.975 and 0.980, respectively. Both consistencies are greater than 0.970. The solution coverage and solution consistency of c_ppp* ~ c_pop*c_phone OR c_cpi*c_ppp*c_phone are 0.843 and 0.974, respectively. Table 2 Three-year lag results from fsQCA.
In the year patterns of 2004–2007 and 2005–2008, there is an antecedent combination, c_cpi*c_ppp*c_phone, which supports proposition P3. Highly-developed countries and low corruption → high ICT development. Table 2 also presents other antecedent combinations, which show the causal complexities of ICT development. Table 3 lists the analyses using MRA. All the adjusted R2 values are greater than 0.880, which shows that MRA has good model fitting capability. All terms have positive impacts on the Internet; however, not all terms are significant in each year pattern. 4.2. Predictive validities To demonstrate the predictive validities using fsQCA, this study performs the prediction using a 3-year lag analysis. This study predicts the rest of the year patterns by using the year pattern for 2004–2007: c ppp c pop c phone OR c cpi c ppp c phone
Raw coverage
Unique coverage
Consistency
Internet(2007) = f(phone(2004), pop(2004), ppp(2004), cpi(2004)) c_ppp* ~ c_pop*c_phone 0.601 0.055 0.975 c_cpi*c_ppp*c_phone 0.788 0.242 0.980 Solution coverage: 0.843 Solution consistency: 0.974 Internet(2008) = f(phone(2005), pop(2005), ppp(2005), cpi(2005)) c_ppp* ~ c_pop*c_phone 0.613 0.050 0.967 c_cpi*c_ppp*c_phone 0.812 0.248 0.972 Solution coverage: 0.861 Solution consistency: 0.966 Internet(2009) = f(phone(2006), pop(2006), ppp(2006), cpi(2006)) c_ppp*c_phone 0.862 0.862 0.972 Solution coverage: 0.862 Solution consistency: 0.972 Internet(2010) = f(phone(2007), pop(2007), ppp(2007), cpi(2007)) c_ppp*c_phone 0.854 0.323 0.977 c_cpi*c_pop*c_phone 0.541 0.010 0.964 Solution coverage: 0.864 Solution consistency: 0.965 Internet(2011) = f(phone(2008), pop(2008), ppp(2008), cpi(2008)) c_cpi* ~ c_pop*c_phone 0.580 0.020 c_ppp* ~ c_pop*c_phone 0.619 0.039 c_cpi*c_ppp*c_pop 0.525 0.107 ~c_cpi* ~ c_ppp*c_pop*c_phone 0.320 0.022 Solution coverage: 0.789 Solution consistency: 0.956
0.978 0.978 0.972 0.952
Table 4 lists the results of the prediction. The consistency of each antecedent combination is greater than 0.960. The solution consistency of each year pattern is also greater than 0.960 and the solution coverage is greater than 0.820. All of these findings show that fsQCA possesses great predicting capabilities. Similarly, this study also uses the equation from the year pattern 2004–2007 to predict the rest of the year patterns by using MRA: Internet = −6.038 + 0.322*phone + 0.002*pop + 0.000*ppp + 1.680*cpi. Table 5 shows the predictive validities using MRA. None of the adjusted R2 values are greater than 0.015 and none of the items in each year is of significance, which shows that MRA is poor in terms of prediction. 4.3. Improvement in ICT development To understand the impacts of antecedent combinations on the improvement in ICT development, this study uses both fsQCA and MRA to perform analyses for the differences in the 3-year lag. The first analysis is
Internetð2010Þ−Internetð2007Þ ¼ fðphoneð2007Þ–phoneð2004Þ; popð2007Þ–popð2004Þ; pppð2007Þ–pppð2004Þ; cpið2007Þ–cpið2004ÞÞ:
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
751
Table 3 Three-year lag results from MRA. Model summary (2004–2007) Model
R
1 .949a a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.901
.897
3.879
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
14231.000 1564.770 15795.000
3557.663 15.046
236.450
b0.000 b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −6.038 phone 1 0.322 pop 1 0.002 ppp 1 0.000 cpi 1 1.680 a. Dependent variable: Internet Internet = −6.038 + 0.322*phone + 0.002*pop + 0.000*ppp + 1.680*cpi
SE
t-Value
Sig Pr N |t|
VIF
0.983 0.039 0.000 0.000 0.350
−6.140 8.300 4.560 1.630 4.800
b0.000* b0.000* b0.000* 0.106 b0.000*
0.000 4.321 1.092 4.069 4.366
Model summary (2005–2008) Model
R a
1 .941 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.886
.881
4.038
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
13121.000 1695.999 14817.000
3280.341 16.308
201.150
b0.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −4.549 phone 1 0.388 pop 1 0.000 ppp 1 0.000 cpi 1 1.146 a. Dependent variable: Internet Internet = −4.549 + 0.388*phone + 0.000*pop + 0.000*ppp + 1.146*cpi
SE
t-Value
Sig Pr N |t|
VIF
0.986 0.040 0.000 0.000 0.352
−4.610 9.630 0.260 1.930 3.260
b0.000* b0.000* 0.796 0.056 0.002*
0.000 3.923 1.102 4.168 4.152
Model summary (2006–2009) Model
R a
1 .944 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.892
.888
4.002
ANOVAa Model
DF
Regression 4 Residual 109 Total 113 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
14405.000 1745.467 16150.000
3601.131 16.013
224.880
b0.000b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop
1 1 1
−3.753 0.391 0.000
0.961 0.039 0.000
−3.91 10.03 0.57
0.000* b0.000* 0.567
0.000 3.932 1.100 (continued on next page)
752
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
Table 3 (continued) Coefficientsa Independent variables
DF
Coefficient
ppp 1 0.000 cpi 1 0.895 a. Dependent variable: Internet Internet = −3.753 + 0.391*phone + 0.000*pop + 0.000*ppp + 0.895*cpi
SE
t-Value
0.000 0.337
Sig Pr N |t|
2.81 2.66
VIF
0.006* 0.009*
4.582 4.068
Model summary (2007–2010) Model
R a
1 .956 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.914
.911
3.689
ANOVAa Model
DF
Regression 4 Residual 107 Total 111 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
15517.000 1456.313 16973.000
3879.286 13.610
285.020
b.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −4.416 phone 1 0.485 pop 1 0.000 ppp 1 0.000 cpi 1 0.997 a. Dependent variable: Internet Internet = −4.416 + 0.485*phone + 0.000*pop + 0.000*ppp + 0.997*cpi
SE
t-Value
Sig Pr N |t|
VIF
0.918 0.035 0.000 0.000 0.320
−4.81 14.06 0.70 1.46 3.12
b0.000* b0.000* 0.483 0.147 0.002*
0.000 3.408 1.085 3.568 4.059
Model summary (2008–2011) Model
R a
1 .948 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.899
.895
3.923
ANOVAa Model
DF
Regression 4 Residual 93 Total 97 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
12789.000 1431.466 14220.000
3197.128 15.392
207.710
b.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 −4.292 phone 1 0.534 pop 1 0.000 ppp 1 0.000 cpi 1 0.956 a. Dependent variable: Internet Internet = −4.292 + 0.534*phone + 0.000*pop + 0.000*ppp + 0.956*cpi
The second analysis is Internetð2011Þ‐Internetð2008Þ ¼ fðphoneð2008Þ–phoneð2005Þ; popð2008Þ–popð2005Þ; pppð2008Þ–pppð2005Þ; cpið2008Þ–cpið2005ÞÞ:
In the first table of Table 6, fsQCA suggests the following three antecedent combinations: c_d_cpi*c_d_ppp*c_d_pop* ~ c_d_phone, c_d_cpi*c_d_ppp* ~ c_d_pop*c_d_phone, OR ~
SE
t-Value
Sig Pr N |t|
VIF
0.958 0.035 0.000 0.000 0.285
−4.480 15.450 0.160 0.700 3.360
b0.000* b0.000* 0.871 0.484 0.001*
0.000 0.412 0.935 0.372 0.417
c_d_cpi*c_d_ppp*c_d_pop*c_d_phone with consistency of 0.916, 0.919, and 0.926, respectively. The solution consistency is 0.904. There is only one antecedent combination in the second analysis, ~c_d_cpi*c_d_ppp*c_d_pop*c_d_phone. When this study examines the recurrent antecedent combinations, there is only one antecedent combination in both analyses: c d cpi c d ppp c d pop c d phone which does not support P4.
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
753
However, there is an antecedent combination in the first analysis:
Table 4 Predictive validities by fsQCA. Raw coverage
Unique coverage
Consistency
Internet(2008) = f(phone(2005), pop(2005), ppp(2005), cpi(2005)) c_ppp* ~ c_pop*c_phone 0.613 0.050 0.967 c_cpi*c_ppp*c_phone 0.812 0.248 0.972 Solution coverage: 0.861 Solution consistency: 0.966 Internet(2009) = f(phone(2006), pop(2006), ppp(2006), cpi(2006)) c_ppp* ~ c_pop*c_phone 0.618 0.047 0.974 c_cpi*c_ppp*c_phone 0.798 0.226 0.983 Solution coverage: 0.845 Solution consistency: 0.977 Internet(2010) = f(phone(2007), pop(2007), ppp(2007), cpi(2007)) c_ppp* ~ c_pop*c_phone 0.620 0.040 0.980 c_cpi*c_ppp*c_phone 0.799 0.219 0.981 Solution coverage: 0.839 Solution consistency: 0.978 Internet(2011) = f(phone(2008), pop(2008), ppp(2008), cpi(2008)) c_ppp* ~ c_pop*c_phone 0.628 0.067 0.978 c_cpi*c_ppp*c_phone 0.760 0.200 0.986 Solution coverage: 0.827 Solution consistency: 0.980
c d cpi c d ppp c d pop c d phone
which supports P4. These two antecedent combinations are rather different. The former one increases in both corruption and population density, while the latter one decreases in both corruption and population density. However, contrary cases exist in the same year with high consistencies. This demonstrates that fsQCA can present contrary cases at the same time to explain the same outcome. The third antecedent combination in the first analysis is c_d_cpi*c_d_ppp*c_d_pop* ~ c_d_phone, which represents an improvement toward a rich country, an increase in population density, but a decrease in corruption and phone lines. Table 7 shows the analyses of differences using MRA. As in both tables in Table 7, both d_phone and d_ppp offer positive impacts but d_ppp and d_cpi negative impacts on the improvement in the Internet. However, the only significant variable is pop in the first analysis and ppp in the second analysis, respectively.
Table 5 Predictive validities by MRA. Model summary (2008) Model
R a
1 .205 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.042
.005
11.131
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
563.660 12885.000 13449.000
140.915 123.893
1.140
0.343b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop ppp cpi a. Dependent variable: Internet
1 1 1 1 1
4.696 −0.044 −0.001 −0.000 1.809
2.820 0.111 0.001 0.000 1.006
1.670 −0.390 −0.810 −0.700 1.800
0.099 0.696 0.420 0.486 0.075
0.000 4.322 1.093 4.070 4.366
Model summary (2009) Model
R a
1 .071 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.005
−0.033
12.084
ANOVAa Model
DF
Regression 4 Residual 104 Total 108 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
77.762 15186.000 15264.000
19.440 146.018
0.130
0.970b
Coefficientsa Independent variables
DF
Intercept phone
1 1
Coefficient 12.793 0.045
SE 3.062 0.121
t-Value
Sig Pr N |t|
VIF
4.180 0.370
b0.000 0.713
0.000 4.322 (continued on next page)
754
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
Table 5 (continued) Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
pop ppp cpi a. Dependent variable: Internet
1 1 1
0.000 −0.000 −0.611
0.001 0.000 1.092
0.340 −0.060 −0.560
0.733 0.951 0.577
VIF 1.093 4.070 4.367
Model summary (2010) Model
R a
1 .152 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.023
−0.017
12.090
ANOVAa Model
DF
Regression 4 Residual 98 Total 102 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
337.406 14325.000 14662.000
84.351 146.170
0.580
0.680b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop ppp cpi a. Dependent variable: Internet
1 1 1 1 1
14.652 0.131 0.000 −0.000 −1.274
3.171 0.122 0.001 0.000 1.117
4.620 1.070 0.290 −0.350 −1.140
b0.000 0.285 0.774 0.724 0.257
0.000 4.223 1.092 4.092 4.370
Model summary (2011) Model
R a
1 .244 a. Predictor: (constant), cpi, pop, ppp, phone
R2
Adjusted R2
Std. error of the estimate
.059
0.015
12.037
ANOVAa Model
DF
Regression 4 Residual 84 Total 88 a. Dependent variable: Internet b. Predictor: (constant), cpi, pop, ppp, phone
Sum of squares
Mean square
F
Sig (Pr N F)
767.162 12170.000 12937.000
191.791 144.881
1.320
0.268b
Coefficientsa Independent variables
DF
Coefficient
SE
t-Value
Sig Pr N |t|
VIF
Intercept phone pop ppp cpi a. Dependent variable: Internet
1 1 1 1 1
12.591 −0.095 −0.002 0.000 −0.869
3.549 0.142 0.001 0.000 1.277
3.550 −0.670 −1.050 1.800 −0.680
0.001 0.504 0.297 0.076 0.498
0.000 4.654 1.139 6.830 4.764
5. Discussion and conclusion This study establishes configural theory for ICT development using fsQCA and compares the capabilities of fsQCA and MRA. The fsQCA results support three propositions: the highly-developed countries, the highly-developed countries with low population density, and the highly-developed countries with low corruption are the sufficient conditions for high ICT development. Fig. 2 depicts the possible relationships. FsQCA shows the causal complexities for the same outcome, while MRA is good at model fitting but falls short in terms of presenting the causal complexities.
This study separates the data and uses the first year pattern to obtain the relationships and then to predict the patterns for the rest of the years. FsQCA shows very strong capabilities in prediction. On the contrary, MRA predicts poorly. As for the improvement in ICT development, fsQCA finds that the improvement toward developed countries and increases in both population density and corruption are the sufficient condition for the improvement in ICT development for both analyses which do not support the proposition. Interestingly, fsQCA also finds a contrary case which is also a sufficient condition for the improvement in ICT development: the improvement toward developed countries and the decrease
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
755
Table 6 Improvement analyses fsQCA. Raw coverage
Unique coverage
Consistency
Internet(2010) − Internet(2007) = f(phone(2007) – phone(2004), pop(2007) – pop(2004), ppp(2007) – ppp(2004), cpi(2007) – cpi(2004)) c_d_cpi*c_d_ppp*c_d_pop* ~ c_d_phone 0.264 0.047 0.916 c_d_cpi*c_d_ppp* ~ c_d_pop*c_d_phone 0.322 0.093 0.919 ~c_d_cpi*c_d_ppp*c_d_pop*c_d_phone 0.255 0.058 0.926 Solution coverage: 0.429 Solution consistency: 0.904 Internet(2011) − Internet(2008) = f(phone(2008) – phone(2005), pop(2008) – pop(2005), ppp(2008) – ppp(2005), cpi(2008) – cpi(2005)) ~c_d_cpi*c_d_ppp*c_d_pop*c_d_phone 0.298 0.298 0.911 Solution coverage: 0.298 Solution consistency: 0.911
Fig. 2. Possible relationships between the antecedent combinations and the outcome.
Table 7 Improvement analyses by MRA. Model summary Internet(2010) − Internet(2007) = f(phone(2007) – phone(2004), pop(2007) – pop(2004), ppp(2007) – ppp(2004), cpi(2007) – cpi(2004)) Model
R2
R
Adjusted R2
Std. error of the estimate
1 .645a 0.416 0.382 2.687 a. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
References
ANOVAa Model
DF
Sum of squares
Mean square
Regression 4 354.251 88.563 Residual 69 497.940 7.217 Total 73 852.191 a. Dependent variable: d_Internet b. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
F
Sig (Pr N F)
12.272
b.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 2.233 phone 1 0.088 pop 1 −0.031 ppp 1 0.000 cpi 1 0.289 a. Dependent variable: Internet
SE
t-Value
Sig Pr N |t|
VIF
0.434 0.085 0.005 0.000 0.688
5.140 1.040 −6.550 0.680 0.420
b0.000 0.301 b0.000* 0.499 0.676
0.000 1.041 1.123 1.114 1.034
Model summary Internet(2011) − Internet(2008) = f(phone(2008) – phone(2005), pop(2008) – pop(2005), ppp(2008) – ppp(2005), cpi(2008) – cpi(2005)) Model
R2
R
Adjusted R2
Std. error of the estimate
a
0.344 0.308 1 .587 a. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
4.464
ANOVAa Model
in both population density and corruption. This antecedent combination supports Proposition 4. FsQCA can show the contrary cases to explain the same outcome while MRA is incapable of doing so.
DF
Sum of squares
Mean square
Regression 4 753.269 188.317 Residual 72 1434.799 19.928 Total 76 2188.068 a. Dependent variable: d_Internet b. Predictor: (constant), d_cpi, d_pop, d_ppp, d_phone
F
Sig (Pr N F)
9.450
b0.000b
Coefficientsa Independent variables
DF
Coefficient
Intercept 1 3.382 phone 1 0.071 pop 1 −0.035 ppp 1 0.000 cpi 1 −0.727 a. Dependent variable: Internet
SE
t-Value
Sig Pr N |t|
VIF
0.660 0.155 0.006 0.000 0.695
5.130 0.460 −5.890 1.450 −1.050
b0.000 0.647 b0.000* 0.153 0.299
0.000 1.027 1.043 1.035 1.023
Balaban, M. E. (2012). Evaluation of human development index and ICT development index, comparative analysis of the OECD and the European members and Turkey. International Journal of Business Research, 12(6), 126–130. Charoensukmongkol, P., & Moqbel, M. (2014). Does investment in ICT curb or create more corruption? Across-country analysis. Public Organization Review, 14(1), 51–63. DeMaagd, K. (2009). The myth of population density and ICT infrastructure. HICSS '09. 42nd Hawaii International Conference on System Sciences (pp. 1–10). Huarng, K. -H. (2010). Essential research in technology management. Journal of Business Research, 63(5), 451–453. Huarng, K. -H. (2011). A comparative study to classify ICT developments by economies. Journal of Business Research, 64(11), 1174–1177. Huarng, K. -H., & Yu, T. H. -K. (2011). Internet software and services: Past and future. Service Industries Journal, 31(1–2), 79–89. Huarng, K. -H., & Yu, T. H. -K. (2014). A new quantile regression forecasting model. Journal of Business Research, 67(5), 779–784. Hudson, H. E. (1995). Access to telecommunications in the developing world: Ten years after the Maitland report. In G. W. Brock (Ed.), Toward a Competitive Telecommunication Industry: Selected Papers From the 1994 Telecommunications Policy Research Conference (LEA Telecommunications Series) (pp. 235–248). Routledge. Hudson, H. E. (2000). Global information infrastructure: Eliminating the distance barrier. In M. Lucas (Ed.), Understanding business: Environments (pp. 235–244). International Telecommunication Union (ITU) (2002). World telecommunications development report. Investopedia (2014). Developed economy. http://www.investopedia.com/terms/d/ developed-economy.asp Ionescu, L. (2013). The positive effect of ICT infrastructure in reducing corruption and increasing transparency. Economics, Management, and Financial Markets, 2, 167–172. Kasikitsakunphon, S., & Vanijja, V. (2013). Factors influencing the intention to use VoIP service by consumers in Thailand. 2013 International Conference on ICT Convergence (ICTC) (pp. 997–1002). Kim, S. -H., & Huarng, K. -H. (2011). Winning strategies for innovation and hightechnology products management. Journal of Business Research, 64(11), 1147–1150. Mehmood, B., & Azim, P. (2013). Does ICT participate in economic convergence among Asian countries: Evidence from dynamic panel data model. Informatica Economică, 17(2), 7–16. Raghuprasad, K. P., Devaraja, S. C., & Gopala, Y. M. (2013). An analysis of knowledge level of farmers on utilization of ICT tools for farm communication. Journal of Rural Development, 32(3), 301–310. Ragin, C. (2008). Redesigning social inquiry: Fuzzy sets and beyond. Chicago: Chicago University Press. Transparency International (2014). Corruption Perceptions Index, Overview. http://www. transparency.org/research/cpi/overview Woodside, A. G. (2013). Moving beyond multiple regression analysis to algorithms: Calling for a paradigm shift from symmetric to asymmetric thinking in data analysis and crafting theory. Journal of Business Research, 63, 463–472. Woodside, A. G., Camacho, A. R., & Lai, W. -H. (2013). Sense making, dilemma, and solutions in strategic management. International Journal of Business and Economics, 12(2), 91–95.
756
K.-H. Huarng / Journal of Business Research 68 (2015) 748–756
Woodside, A. G., Schpektor, A., & Xia, X. (2013). Triple sense-making of findings from marketing experiments using the dominant variable-based logic, case-based logic, and isomorphic modeling. International Journal of Business and Economics, 12(2), 131–154. Woodside, A. G., & Zhang, M. (2013). Cultural diversity and marketing transactions: Are market integration, large community size, and world religions necessary for fairness in ephemeral exchanges? Psychology & Marketing, 30(3), 263–276.
World Bank (2014). GNI per capita, PPP (current international $). http://data.worldbank. org/indicator/NY.GNP.PCAP.PP.CD Yu, T. H. -K. (2011). Heterogeneous effects of different factors on global ICT adoption. Journal of Business Research, 64(11), 1169–1173.