Snapshots of a state of flux: how Brazilian industrial firms differ in the adoption of digital technologies and policy implications

ABSTRACT

This article analyses the adoption of digital technologies by a panel of Brazilian industrial firms in time. An ordered logistic regression model was used to relate modes of adoption to firm size, sector, capabilities and readiness levels to prepare for the future. Digitalisation levels are very basic in 2017. By 2027, most firms expect to move forward. Regardless of size and sector, more or less advanced modes of adoption are associated with higher or lower levels of capabilities and readiness. Two policy implications: policy must aim “moving targets” and be flexible enough to fosters firms at various stages of development.

KEYWORDS:

• Technology adoption
• digital technologies
• industrial firms
• Brazilian industry
• industrial policy

JEL CLASSIFICATION:

• O14
• 025
• O33
• M21
• L16
• L60

Acknowledgements

We thank Henrique Schmidt Reis for his valuable research assistance.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Material

Supplemental data for this article can be accessed here.

Notes

1. This survey is part of the project “Indústria 2027: Riscos e Oportunidades para o Brasil diante de Inovações Disruptivas”, contracted by the Brazilian Industrial Board (CNI) to the Instituto de Economia, Universidade Federal do Rio de Janeiro and the Instituto de Economia, Universidade Estadual de Campinas (www.portaldaindustria.com.br/cni/canais/industria-2027).

2. The consequences of digital technologies are more complex when these are blended in systems, products and processes with other pervasive technologies – nanotechnologies, genomics, advanced materials, among others- (OECD 2017; IEL 2018).

3. 2010 prices. Source: Brazilian Institute of Geography and Statistics (IBGE).

4. OECD (stats.oecd.org) and Brazilian Innovation Survey (ibge.gov.br/home/estatistica/economia/industria/pintec).

5. The questionnaire reflects a 30-year experience in similar exercises by the research team (ie.ufrj.br/index.php/gic/home).

6. CNI Research is well experienced in surveys in subjects like industrial activity, foreign trade, employment, business confidence and investment trends (www.portaldaindustria.com.br/estatisticas).

7. Agroindustries: food products, beverages and tobacco; Automotive: motor vehicles and auto parts; Basic Metals: iron, steel, pulp, cement; Capital Goods: electrical machinery, machinery and equipment; Chemicals: petrochemicals, rubber and plastic products; Consumer Goods: textiles, garments, footwear, durable goods; ICT: office and computing machinery, communication instruments (software not included).

8. The six-digit level of occupation categorisation of the Brazilian Classification of Occupations (CBO) – compatible with the International Standard Classification of Occupation (ISCO)-was used. From 2,614 occupations, 174 were selected including researchers and poly-scientific professionals; professionals of natural sciences, physics and engineering; professionals from biological sciences and similar areas.

9. Moses Abramovitz (1986) seminal work is the source of inspiration.

10. The intensity of movement was obtained for each organisational function, of each company, simply by multiplying 2017 position (1, 2, 3 or 4) by the 2027 position (also 1, 2, 3 or 4).

11. Being a generic function, the fifth function, business management, was put aside for the econometric exercise.

12. The generic function “business management” was also put aside for the readiness variable.

13. A comprehensive review of regressions with categorical variables is in Agresti (1996) and Agresti (2002).

14. Besides the proportional and the partial proportional odds model, Abreu, Siqueira, and Caiaffa (2009) suggest two other options: the ordered continuous ratio and the stereotype logistic model. The first is adequate when a special interest exists in a specific category of the response variable. The second is indicated when the ordinal outcome variable does not come from an aggregated continuous variable.

15. The ordered logistic regression (proportional odds model) is a special case of the generalised ordered logistic (partial proportional model). For an outcome with M categories, the parallel model is written as:

$P\left(Y_i>j\right)=\frac{\mathrm{e}\mathrm{x}\mathrm{p}\left({\alpha }_j+X_i{\beta }_j\right)}{1+\left[\mathrm{e}\mathrm{x}\mathrm{p}\left({\alpha }_j+X_i{\beta }_j\right)\right]},j=1,2,\dots ,M-1$

M-1 equations are simulated simultaneously with M-1 coefficients. For the specific case of ordered logistic (proportional odds), the betas are the same for each j.

16. Differences among sectors may exist. The analysis will associate modes of evolution to the probability of firms belonging to specific sectors.

17. The command line in the STATA software was: gologit GER_TIPOLOGIA_CLASSIF ATITUD_TIPOLOGIA_CLASSIF COD_PORTE_RAIS2i.SISTEMA_PRODUTIVO4 FAIXA_QUALIF_RAIS brant

18. For more details about Brant Test see Long and Freese (2014).

19. The command line was: gologit2 GER_TIPOLOGIA_CLASSIF ATITUD_TIPOLOGIA_CLASSIF COD_PORTE_RAIS2 i.SISTEMA_PRODUTIVO4 FAIXA_QUALIF_RAIS, autofit lrf. The command “autofit” automatically performs the Wald Test and selects all independent variables that confirm the proportional odds assumption.

Additional information

Funding

The research from which this article was derived from was funded by the Instituto Euvaldo Lodi (IEL) of the Confederação Nacional da Indústria (CNI), Brazil.