Discussion of Limitations

The hypothesis that accounting performance differs in regard to the extent of digitalization was rejected. However, the results are subject to certain limitations. To ensure a holistic perspective on limitations, both internal, external, and construct validity will be discussed along with the reliability of the study.

7.2.1 Internal validity

The instrumentation, even though bringing objectivity, entailed some difficulties due to having multiple sources for the data collected. A further limitation involves the sample size being relatively small due to limited/ missing data points concerning stock and flow variables. A limited sample size can be prone to extreme values (Saunders et al., 2009) that in turn can distort the entire data set and impacts the opportunity to establish a causation for the model results. Additionally, the lack of an existing proven digital maturity assessment framework prohibits the possibility of establishing a causal relationship between the selected maturity framework and firm accounting performance.

Moreover, the geographical focus on Europe may impose an additional limitation. Since Europe was chosen for reasons of coherence it cannot be excluded that European banks may be more or less digitally evolved than American, Asian or Australian financial institutions. The mentioned factors can altogether affect the internal validity of the findings.

95 7.2.2 Construct validity

This paper used stock and flow variables as a proxy for digitalization. Furthermore, accounting performance measure were used as a proxy for firm performance. This section discusses whether stock and flow variables in fact represent the extend of digital transformation. Moreover, it will be discussed if accounting performance measure can account for relevant aspects of firm performance.

Previous studies (Decarolis & Deeds, 1999; Miranda et al., 2011; Roper & Hewitt-Dundas, 2015) have established evidence for stock and flow variables being a valid proxy for knowledge capability and maturity. Hence, applying this concept on digitalization may be regarded as adding value to the overall construct validity.

Nonetheless, the lack of an existing digital maturity framework based on financial data or mere objectively obtained data (Reis et al., 2018) poses a challenge on creating a sound framework of measurement. Therefore, the constructed measurement for digital maturity may not capture all required aspects since, for instance, soft measures such as the existence of an aligned digital strategy, supporting governance mechanisms, staff trainings and organizational culture were not incorporated in the measurement.

Moreover, firms have a voluntary (unregulated) choice to report intangible assets (Bughin &

Manyika, 2013; Saunders & Brynjolfsson, 2016) which may result into inferior construct validity as the measured stock of intangible asset may not depict the actual asset stock. Inferencing that the digital maturity measure may not be valid concerning the stock values. Wyatt (2005), however, argues that established firms have a strong incentive to provide information about unobservable assets when projects such as digital transformation initiatives advance as associated intangibles assets tend to become embodied in other assets and become therewith less risky. Furthermore, firms have incentives to report intangible assets where management expects that those intangibles will lead to superior, future performance (Wyatt, 2005). Therefore, firms with more advanced digital projects may record higher levels of intangibles. This view is aligned with this research assumption that higher levels of digital asset stock demonstrate higher levels of digital maturity.

96

While this paper focuses on accounting performance, it is known that no single set of performance measure can account for all aspects of firm performance. Even though firm performance has been assessed using a variety of accounting measures, there is no universal guide for the most appropriate choice. Previous studies (Floyd & Wooldridge, 1990; Becalli, 2007; Sujud & Hashem, 2007; Miranda et al., 2011; Muiruri & Ngari, 2014, Akhisar et al., 2015; Scott et al., 2017) have established evidence for the performance measures used, both, within the banking industry and outside. Hence, applying this selection of accounting measures may be regarded as adding value to the construct validity.

It is worthwhile mentioning that performance and digitalization was measured objectively. The concept of digital transformation is not a uniform process which makes it difficult to distinguish in different advancement levels. Therefore, it can be argued that choosing a statistical clustering method such as k-means clustering to divide into digital degrees may be a viable and valid approach to help moderate the potential for bias.

7.2.3 External validity

The main concern in the discussion about external validity is the issue of conducting the investigation in a single industry and large-scale organizations. The limitation involves the representativeness of the sample. Any inferences based on the results are restricted to the nature of the banks. The sample consists of banks being the largest European banks measured in market capitalization.

Moreover, the financial market is imposed to strict regulation which is likely to influence the results obtained to different times. Furthermore, financial institutions, like all public companies, are obliged to adhere to certain accounting standards that are subject to change and adjustments.

Naturally, the implementation of certain measures may differ between organizations of different size and industry, therefore, the application of this model to other practices has potential, it may, however, be that the results obtained in this paper are difficult to generalize well for future periods.

97 7.2.4 Reliability

Reliability refers to the degree to which researchers will arrive at the same insights if a study was to be conducted along the same steps as this paper presents. Although the analysis of this paper is based on secondary data which is publicly available, there has been, a challenge concerning the completeness of the dataset obtained. However, each step of data collection, preparation and transformation is described in a transparent manner and therewith reliable. Nevertheless, ratio computation was done manually and could, thus, be prone to the errors in reporting. Moreover, it is questionable if a researcher would arrive at the same digital maturity groups, even though the groups were clustered automatically by means of k-means clustering. The labelling of the group levels (beginning/norming, transforming and maturing) as well as the number of levels assigned is sensitive to subjectivity and interpretation of the stock and flow values displayed in each cluster.