The model used by Brynjolfsson and Hitt was quite simple, they imposed the following general form in relating the firms quantity of output produced to inputs used:
Q is the quantity of output
F is the production function
C is the computer capital input
K is non-capital input
S is information systems staff labour
L other labour and expenses
i is the specific firm
t is the time interval
They assumed that the production function conforms to the Cobb-Douglas specification
β1 is the output elasticity of computer capital
β3 is the output elasticity of information systems staff labour
Since regression models need to be linear in the parameters a logarithmic transform is done to create the log-log linear model:Where: Qit = output of a firm in industry i in year t
Cit = computer capital
Kit = non-computer capital
Sit = information systems staff labour capital
Lit = other labour and expenses
β is a vector of parameters to be estimated
Log denotes the natural logarithm
ε is a vector of random variables
The Brynjolfsson and Hitt paper provided evidence for IT made a “substantial and statistically significant contribution to the output of firms”. This was in contrast to the evidence of previous papers which supported the notion of the “productivity paradox” existing that is that “despite enormous improvements in the underlying technology, the benefits of IS spending have not been found in aggregate output statistics”. Brynjolfsson and Hitt attributed the differences in the results owing to the following factors:
- Measurement problems associated with the use of industry level and economic level data.
- Data being used that had not incorporated the impact IT due to the lag effect and time needed for learning and adjustment
- The intangible benefits that IT produces not being incorporated correctly into the estimated models
All of the fore mentioned issues would result in situation where one of the assumptions of classical linear regression model which that the regressors and disturbances are independent:
X is a regressor such as computer capital
ε is a disturbance term, random variable
When this fundamental assumption is violated it will result in biased, inefficient and inconsistent estimates of our parameters in our regression model and consequently the incorrect conclusions being drawn form the hypothesis tests.
- Using firm level data which is more recent and incorporates the fact that firms have undertaken restructuring of their business process and are realising the benefits ofIT
- Using iterated seemingly unrelated regression model (ISUR) to improve the efficiency of their estimates as it can directly address serial correlation (relationship between different firms in the same time period) and missing observations.
- Using instrument variables to control for omitted variable bias and the violation of the assumption that regressors and the disturbances term are uncorrelated via three stage least squares (3SLS) estimation technique.