Effects of firm and IT characteristics on the value of e-commerce initiatives: An inductive theoretical framework

Abstract

We explore the theoretical foundations on how firm and IT characteristics explain the market value variations in e-commerce initiatives by examining the announcements of 946 e-commerce initiatives in the public media. Our approach combines the Event study methodology and Decision tree induction to examine the main and interaction effects of IT and firm characteristics on Cumulative Abnormal Returns (CAR). In particular, we generate complex interaction models that can guide e-commerce investment decisions so managers can know, for example, which combination of IT and firm characteristics are more likely to be viewed positively by investors. The selected study variables as well as explanation of the proposed framework are informed by innovation, resource-based view, transaction cost economics and complementarity theories. We have inductively developed a set of propositions that can be deductively tested to assess the validity of our proposed theoretical framework. Hence our study provides an initial roadmap for theory development on e-commerce and CAR.

Appendices

Appendix A1: Sample e-commerce announcement and classification

Appendix A2: Sample e-commerce announcement and classification

Appendix B. Hypothesis abduction and evaluation

3.1 First-order sibling rules hypothesis

Consider a pair of sibling rules presented in Table 6 (generated from the DT induction) where all conditions are the same (Innovativeness is Transformational) except for the one involving the given discriminating variable (e.g. Governance):

• IF Innovativeness is Transformational & Governance is Unilateral THEN CAR is Positive with probability 74.7% and N (i.e. Number of Cases) = 115;

• IF Innovativeness is Transformational & Governance is Joint THEN CAR is Positive with probability 84.3% and N = 97.

The existence of this pair of sibling rules leads to the creation of the hypothesis: “IF Innovativeness is Transformational THEN Governance is a predictor of CAR.” Governance is a discriminating predictor in this case.

For the given target event (e.g. CAR is Positive), the posterior probabilities for each sibling node are compared. If for any pair of sibling nodes, the relevant posterior probabilities are very different, then this would suggest that the given variable is a predictor for the target event (Osei-Bryson and Ngwenyama 2004). In this manner, a given set of sibling rules can be used to generate and test hypotheses that involve conjecturing that the given variable is a predictor of CAR. If the number of cases associated with a given set of sibling nodes is sufficiently large, then the hypothesis may be subjected to statistical analysis. The statistical test used here is difference of proportion test to confirm that the difference in posterior probabilities (proportions or relative frequencies of the abnormal events) for the sibling nodes of the discriminating variable did not occur by chance. The difference is between two proportions (p1 and p2) based on two independent samples of size n ₁ and n ₂ with sample proportions ${\hat{P}_1}$ and ${\hat{P}_2}$.

According to (Groebner et al., 2008), the test statistic for the difference of proportion test is given by:

$$ Z = \frac{{{{\hat{P}}_1} - {{\hat{P}}_2}}}{{\sqrt {{\frac{{{{\hat{P}}_1}\left( {1 - {{\hat{P}}_1}} \right)}}{{{n_1}}} + \frac{{{{\hat{P}}_2}\left( {1 - {{\hat{P}}_2}} \right)}}{{{n_2}}}}} }} $$

From the sample pair of sibling rules,

$$ \begin{array}{*{20}{c}} {{{\hat{P}}_1} = 0.843,\;{{\hat{P}}_2} = 0.747,\;{n_1} = 115\;{\hbox{and}}\;{n_2} = {97}} \hfill \\{{\hbox{Thus}}\;Z = \frac{{0.843 - 0.747}}{{\sqrt {{\frac{{0.843\left( {0.157} \right)}}{{115}} + \frac{{0.747\left( {0.253} \right)}}{{97}}}} }}} \hfill \\{{\hbox{Thus}}\;Z = \frac{{0.096}}{{\sqrt {{{0}{.001364} + {0}{.001643}}} }}} \hfill \\{Z = 0.096/0.05484 = 1.750425.} \hfill \\{P\left( {\hbox{Z}} \right) = 0.0{4}.} \hfill \\\end{array} $$

Similarly, for the other sibling rule in Table 6,

• IF Innovativeness is Transformational THEN CAR is Positive with probability 77% and N (i.e. Number of Cases) = 379;

• IF Innovativeness is Executional THEN CAR is Positive with probability 36% and N = 567.

$$ \begin{array}{*{20}{c}} {{{\hat{P}}_1} = 0.77,\;{{\hat{P}}_2} = 0.36,\;{n_1} = 379\;{\hbox{and}}\;{n_2} = 567} \hfill \\{{\hbox{Thus}}\;Z = \frac{{0.77 - 0.36}}{{\sqrt {{\frac{{0.77\left( {0.23} \right)}}{{379}} + \frac{{0.36\left( {0.64} \right)}}{{567}}}} }}} \hfill \\{{\hbox{Thus}}\;Z = \frac{{0.41}}{{\sqrt {{{0}{.000467} + {0}{.000406}}} }}} \hfill \\{Z = 0.41/0.029557 = 13.87138} \hfill \\{P\left( {\hbox{z}} \right) = 0.0.} \hfill \\\end{array} $$

Details of the difference of proportion tests performed at the 5% level are presented in Table 11 below.

Table 11 Difference of proportions tests for first-order sibling rules hypotheses

3.2 Second-order sibling rules hypotheses

A first-order sibling rules hypothesis is based on a set of sibling rules. A second-order sibling rules hypothesis is based on two sets of sibling rules (say S₁, S₂) that have the following conditions: (Tables 12, 13, and 14, Fig. 1)

Table 12 Format of second-order sibling rules hypotheses

Table 13 Example of second-order sibling rules hypotheses without moderator

Table 14 Example of second-order sibling rules hypothesis with moderator

Fig. 1

Decision tree example

Similar to the First Order Sibling Rules (Groebner et al. 2008), the difference of proportion testing for the Second-Order Sibling Rules Hypothesis involves computing Z which is given by:

$$ Z = \left( {\left( {{\rho_{11,21}} - {\rho_{11,22}}} \right)-\left( {{\rho_{12,21}} - {\rho_{12,22}}} \right)} \right)/{\hbox{s}}_{\rm{p}}^{1/2} $$

where ${{\hbox{s}}_{\rm{p}}} = \left( {{\rho_{11,21}}\left( {1 - {\rho_{11,21}}} \right)/{{\hbox{n}}_{11,21}} + {\rho_{11,22}}\left( {1 - {\rho_{11,22}}} \right)/{{\hbox{n}}_{11,22}} + {\rho_{12,21}}\left( {1 - {\rho_{12,21}}} \right)/{{\hbox{n}}_{12,21}} + {\rho_{12,22}}\left( {1 - {\rho_{12,22}}} \right)/{{\hbox{n}}_{12,22}}} \right)$.

Using example in Table 12 which is also the rule for H_2.1 (Table 8),

$$ \begin{array}{*{20}{c}} {{\rho_{{11},{21}}} = 0.{69},{\rho_{{11},{22}}} = 0.{43},{\rho_{{12},{21}}} = 0.{82}\;{\hbox{and}}\;{\rho_{{12},{22}}} = 0.{27}} \hfill \\{{{\hbox{n}}_{{11},{21}}} = {134},{ }{{\hbox{n}}_{{11},{22}}} = {315},{ }{{\hbox{n}}_{{12},{21}}} = {245},{\hbox{ and }}{{\hbox{n}}_{{12},{22}}} = {252}} \hfill \\\end{array} $$

$\left( {{\rho_{11,21}} - {\rho_{11,22}}} \right) - \left( {{\rho_{12,21}} - {\rho_{12,22}}} \right) = 0.55--0.26 = 0.29$ (See Table 13, B2B has higher difference than B2C).

$$ \begin{array}{*{20}{c}} {{\hbox{Sp}} = \left( {0.001596 + 0.000778 + 0.000602 + 0.000782} \right) = 0.003759} \hfill \\{{\hbox{S}}{{\hbox{p}}^{1/2}} = 0.06131} \hfill \\{{\hbox{Thus}}\;Z = 0.29/0.06131 = 4.73004.} \hfill \\{P\left( {\hbox{Z}} \right) = 0.000001.} \hfill \\\end{array} $$

Details of difference of proportion tests performed at the 5% significant level for the Second-Order Rules Hypotheses are presented in Table 15 below.

Table 15 Difference of proportion tests for second-order rules hypotheses