The Working-from-Home Natural Experiment in Sydney, Australia: A Theory of Planned Behaviour Perspective

It might be helpful for readers to understand the statistical analysis process and appreciate the contributions to have a clear set of hypothesis statements using the set of key variables (Attitude, Norms, Behavior Control, Opportunity). e.g. H1: A is positively related to B. H2: C is positively related to B. so on and so forth.

Many thanks. In order to more clearly describe the statistical analysis process to the readers, we have now introduced a set of hypotheses that corresponds to these results.

The survey seems to use a categorical (Yes-No) scale. The reporting is a mix of categorical and continuous scales. It is not evident to the readers how the analysis process is done. For instance, Table 2. How are "changes" defined and calculated? 

The survey responses are a mix of binary and ordinal scales. In order to more clearly describe how the analysis process was performed, an extended description has been added to section 2.3 (p8), as well as clarification regarding this.

Table 2 (p9) has also been updated to reflect the analysis process more clearly, and describe how changes are defined and calculated etc.

Section 2.3 now reads as follows:

The statistical analysis was undertaken using the Scientific Python Development Environment (Spyder) version 4.01 through the Anaconda distribution of the Python languages. Methods were chosen based on the following:

•             To test the hypothesis that intentions to WFH have changed as a result of the experiences during COVID lockdowns we use a Chi-Square test to explore the association between two ordinal variables.

•             We use Cronbach’s Alpha to evaluate the internal consistency of the variables included in each of the three constructs that determine intention. In other words, if different measures of the same construct provide similar results. A Cronbach’s Alpha of above 0.7 is acceptable.

•             To explore the relationship between variables, we use the Pearson correlation coefficient. We have a mix of ordinal, and binary data, and therefore Spearman rank correlations could have also been used, which involves an additional transformation before calculating the Pearson coefficient (our coding is a defacto ranking system). When Pearson correlation coefficient is calculated on two binary variables, it is referred to as a Phi coefficient.

•             To evaluate the influence of attitudes, perceived behavioural control and subjective norms, on the intention to WFH, we use Ordinary Least Square regression as it provides a simple linear equation to describe the relationship. We have no evidence to suggest a non-linear relationship between variables, and we choose the simplest possible explanation.

•             To check for issues associated with multi-collinearity between independent variables, we check for Variance Inflation Factors (VIF), as is standard practice in regression modelling.

•             The data from the survey is a mix of ordinal and binary scales, for the most part based on Likert scales. These are in most cases coded to a numeric scale between 0 and 1, with equal spacing based on the ranking for most ordinal scales. Further details on relevant scales are provided in each of the tables.

The appendix (survey questions) is not included in the manuscript. This made it difficult to understand the process and evaluate the statistical analysis.

We are sorry to hear this. We uploaded an appendix containing the survey questions as a separate file and believed that this would be made available to the reviewers.

We will make a comment to the Editor as part of this response, to ensure that the appendix is made available to the reviewers.

1.     Topic originality and relevance to the field

The manuscript deals with the working-from-home (WFH) experience in Sydney during the COVID-19 crisis as seen through the lens of the Theory of Planned Behavior. To be more specific, the article examines the main drivers that determined the Australian workers’ involvement in the WFH procedures and tries to emphasize the influence of this unique experiment upon the self – reported measures of productivity, health and wellbeing indicators.

By employing an in-depth approach on the main drivers determining the intention to continue the WFH practice after the pandemic, the paper performs a comprehensive analysis on workers’ attitudes, subjective norms and the perceived behavioral control, in order to predict the extent to which WFH routines are going to be promoted or disengaged in the post-COVID 19 era.

Many thanks for your kind words and recognition of the originality and relevance of this study.

2.     Literature Review

This section is embedded in the “Introduction” part. I strongly recommend the use of a distinct section of the paper called “Literature Review” in order to put the WFH phenomenon in context.

Within this section of the paper, I believe that highlighting the research gap that authors are trying to fill could significantly improve the process of understanding the motivations that triggered authors’ interest towards specific variables and statistical analyses performed in the paper.

Thank you for the suggestion. We have now restructured the paper accordingly, to include a distinct LR section that highlights the research gap more clearly, and the factors that drive the decision to WFH.

This new section now appears as 1.1 (p3) and is followed by a Knowledge Contribution section 1.2 (p4), which presents a conceptual framing of what drives the intention to WFH.

3.     Methods

The content of this section looks scientifically sound. Nevertheless, I kindly ask the authors to elaborate more on the following statement included in the Abstract: “ A model developed from the insights in this paper has a correlation of 0.9 with the spatial distribution of residential mobility changes”. I think this assertion is not sufficiently supported by the content of the paper.

Many thanks for identifying this error.  We agree that the evidence to support this statement was not provided in the paper.

To rectify this, we have done two things. Firstly, we removed this statement in the abstract (as we think it is too early to provide a full description of the methodology). Secondly, we elaborated further on these activities in section 4.4 (p20) so that more details are provided (although not sufficient for including it in the abstract).

Secondly, I think that some additional information on the sampling procedures would be very helpful for researchers from other parts of the Globe who would be interested in replicating the present study. To be more specific, I suggest authors to provide further details on the approach they employed in order to determine the sample size and to choose the most appropriate sampling method.

Thanks again. In order to provide additional information on the sampling procedures, we have added the following paragraph to Section 2.1 (p5): “The sample size was chosen on the basis of attempting to achieve a representative sample, using the Taro Yamane equation [46], across the region which has a working population of approximately 2,500,000, and a chosen margin of error of 2%. This suggested a sample size of 2,500.”

I would also ask the authors to indicate the software programs (and their versions) which were used in order to carry out statistical investigations such as correlation analyses, least square regression or Cronbach Alpha’s substantiation.

Thanks.  Section 2.3 (p8) has been revised to include a description of the software programs which were used.

Section 2.3 now reads as follows:

The statistical analysis was undertaken using the Scientific Python Development Environment (Spyder) version 4.01 through the Anaconda distribution of the Python languages. Methods were chosen based on the following:

•             To test the hypothesis that intentions to WFH have changed as a result of the experiences during COVID lockdowns we use a Chi-Square test to explore the association between two ordinal variables.

•             We use Cronbach’s Alpha to evaluate the internal consistency of the variables included in each of the three constructs that determine intention. In other words, if different measures of the same construct provide similar results. A Cronbach’s Alpha of above 0.7 is acceptable.

•             To explore the relationship between variables, we use the Pearson correlation coefficient. We have a mix of ordinal, and binary data, and therefore Spearman rank correlations could have also been used, which involves an additional transformation before calculating the Pearson coefficient (our coding is a de-facto ranking system). When the Pearson correlation coefficient is calculated on two binary variables, it is referred to as a Phi coefficient.

•             To evaluate the influence of attitudes, perceived behavioural control and subjective norms, on the intention to WFH, we use Ordinary Least Square regression as it provides a simple linear equation to describe the relationship. We have no evidence to suggest a non-linear relationship between variables, and we choose the simplest possible explanation.

•             To check for issues associated with multi-collinearity between independent variables, we check for Variance Inflation Factors (VIF), as is standard practice in regression modelling.

•             The data from the survey is a mix of ordinal and binary scales, for the most part, based on Likert scales. These are in most cases coded to a numeric scale between 0 and 1, with equal spacing based on the ranking for most ordinal scales. Further details on relevant scales are provided in each of the tables.

4.     Conclusions and Recommendations

The section contains a comprehensive summary of the results of the study. However, I think that authors should make proper linkages to other up-to-date studies conducted in the same field. I suggest authors to propose some future research directions to deepen the knowledge regarding the determinants of WFH propensity.

Thank you for this suggestion. We have now added a number of linkages, to other up-to-date studies conducted in this field, in addition to proposing some potential future research directions (p21-22).

5.     Technical Notes

Last but not least, I consider that the flow of ideas in the manuscript should be improved by correcting a few minor English grammar errors. For instance, see the following phrase from page 3 of your manuscript (lines 138-139): “To explore these issues, we interrogated survey responses both from those who had worked from home, as well as many who hadn’t” or the phrase from lines 163-165: “A summary of the participant demographics, housing and socio-economic distributions are shown in Table 1 and spatial distribution as per Figure 1”.

                Besides, authors should remove the number of the page from the in-text citations which are inserted in the body of the paper (for instance, see lines 32, 35, 82 etc.)

After performing the above-mentioned adjustments, I believe the paper will be ready for publication.

Many thanks, paper has now been thoroughly proof-read, and English grammar errors amended accordingly.

Page numbers need to be included when providing quotes using the ACS reference style, but we have changed our referencing of the page numbers, so that it aligns with the formatting requirements for the journal.