The Psychology of Sharing: Multigroup Analysis among Users and Non-Users of Carsharing

Round 1

Reviewer 1 Report

This article investigates some aspects related to the factors that determine the intention to use car-sharing, through the results of a cross-cultural survey conducted on groups of users and non-users in Italy and Sweden. The paper is well-written, methodologically sound, and the results, while not particularly innovative in the already very rich landscape of studies on intention to use this transport mode, are well presented and argued.

The research questions presented are answered in the numerical results of the modeling approaches considered: the Multigroup Confirmatory Factor Analysis to verify the effects of model paths between groups and in each group and then a structural equation model (SEM) in the framework of TPB and TAM to evaluate the effects of habits, climate morality, personal norms, PBC, perceived usefulness and trust on car-sharing intention of use.

My main comments are not on the methods used, or their accuracy, but rather on some aspects related to the choice of conducting a study on different European countries and the conclusions that can be derived from this structure of the survey.

1) In my opinion, the authors should discuss more robustly the choice of a cross-cultural analysis on samples of users and non-users from Italy and Sweden. Why Italy and Sweden? This remark is not enough (lines 501-504): “Regarding the differences between nationalities, one may suggest that this effect is due to the longer availability of carsharing services in the Swedish cities (dated from 1970), compared to the Italian cities (emerging in 2000) [78, 79]. However, further research on this difference would be illuminating to draw more concise conclusions.”. That's certainly a fair comment, but it's an ex-post deduction based on the results. It would be appropriate to highlight the preliminary assessments that were made to choose samples from these two populations: cultural, historical, behavioral assessments (e.g.: modal choice habits, orientation towards car ownership, gender issues, regulation, policies at the national level to promote sustainable behavior, etc. ...).

2) It does not seem to me that aspects related to orientation towards public transport systems were considered in the questionnaires. In the context of modal choice, carsharing is often included in semi-collective systems, in cooperation or complement with other urban public transport systems. Particularly in the case of Italy, where historic city centers have strong restrictions on the use of private cars, e-car sharing is becoming more and more popular as a modal alternative in modal-chains that involve the use of more than one public transport system (e.g. bus, light rail transit and car sharing for the last mile). More generally, the attitude of users and non-users towards public transport could certainly help to explain more fully some of the results obtained, for example those on climate morality.

Some minor comments and suggestions:

- The illustration of the SEM model results (Sec. 3.2) should be expanded; it seems very concise compared to other parts of the article.

- Bold values in Table 4 and 5 indicate the statistical significance of the predictors considered. In tab. 4, "trust" should probably be in bold for nonusers.

- The "conclusion" section is too concise, especially compared to the one containing the discussions. I suggest expanding it, possibly by transferring some of the considerations in sec. 4 to the next one.

Author Response

1 - In my opinion, the authors should discuss more robustly the choice of a cross-cultural analysis on samples of users and non-users from Italy and Sweden. Why Italy and Sweden? This remark is not enough (lines 501-504): “Regarding the differences between nationalities, one may suggest that this effect is due to the longer availability of carsharing services in the Swedish cities (dated from 1970), compared to the Italian cities (emerging in 2000) [78, 79]. However, further research on this difference would be illuminating to draw more concise conclusions.”. That's certainly a fair comment, but it's an ex-post deduction based on the results. It would be appropriate to highlight the preliminary assessments that were made to choose samples from these two populations: cultural, historical, behavioral assessments (e.g.: modal choice habits, orientation towards car ownership, gender issues, regulation, policies at the national level to promote sustainable behavior, etc. ...).

More information about the choice of countries was added in lines 61-65; 72-77.

2 - It does not seem to me that aspects related to orientation towards public transport systems were considered in the questionnaires. In the context of modal choice, carsharing is often included in semi-collective systems, in cooperation or complement with other urban public transport systems. Particularly in the case of Italy, where historic city centers have strong restrictions on the use of private cars, e-car sharing is becoming more and more popular as a modal alternative in modal-chains that involve the use of more than one public transport system (e.g. bus, light rail transit and car sharing for the last mile). More generally, the attitude of users and non-users towards public transport could certainly help to explain more fully some of the results obtained, for example those on climate morality.

            The groups' descriptive data of the frequency of PT use was added in Table 1 and some information in the intro (see lines 74-76). However, the attitudes towards PT were not the focus of this study, making it difficult to expand the discussion surrounding this topic without motivation from the research questions. We point out in the discussion section that further investigation into PT use could help explain the relationship between concerns to climate change and the use of different modes of shared transportation (see lines 737-739).

Some minor comments and suggestions:

3 - The illustration of the SEM model results (Sec. 3.2) should be expanded; it seems very concise compared to other parts of the article.

            To address the comments from reviewer 2, new χ2 difference tests were performed to measure the null hypothesis that the coefficients are the same across the four groups. Please see the lines 534-549. With these new analyses, section 3.2 was extended, and the results better presented.

4 - Bold values in Table 4 and 5 indicate the statistical significance of the predictors considered. In tab. 4, "trust" should probably be in bold for nonusers.

            In Table 4, the p-value for Trust in the group of non-user is .099. We worked with the 95% confidence level threshold, and therefore, the estimate is not in bold.

 

5 - The "conclusion" section is too concise, especially compared to the one containing the discussions. I suggest expanding it, possibly by transferring some of the considerations in sec. 4 to the next one.

Three paragraphs from the discussion were moved to the conclusion section; see lines 734-737; 739-821.

Reviewer 2 Report

This is an interesting and methodologically-sound article that compares results between two countries, something lacking from the literature on transport and psychological variables. I have relatively minor comments, mostly related to clarifying the text rather than the modeling effort per se.

- I recommend the authors rename the construct "Control" to "Perceived behavioral control" to better align with the TPB and to make the abstract more clear.
- Section 1.3 should explicitly say which theory (TPB or TAM) each latent construct comes from
- When equivalence is computed across groups - is this done across non-users and users, across countries, or both?
- When comparing coefficients for users and non-users in the discussion and Tables 4 and 5, is is possible to do a hypothesis test of the null hypothesis that the coefficients are the same? For most coefficients the CIs overlap, which suggests there may not be a difference, but this is an underpowered test. Furthermore, the authors also compare one coefficient (habit in italy) based on it being significant in one model and not in the other, even though the CIs overlap, which seems inconsistent.
- Similarly, the discussion notes slight differences in coefficients between Italy and Sweden. Is it possible to do a test to see if these differences are statistically significant?
- Line 517 there is a typo - "more less"

Author Response

1 - I recommend the authors rename the construct "Control" to "Perceived behavioral control" to better align with the TPB and to make the abstract more clear.

Control was assessed by a combination of perceived behavior control (from TPB), perceived ease of use, and usefulness (both from TAM). We think that calling the construct “Control” as “Perceived behavioral control” could mislead the reader to interpret it as the classic construct from TPB, which it’s not the case in our study. Therefore, we would like to maintain the construct’s name as it is. Besides, we clarify in lines 297-299; 378-382; 688-690 how this construct was measured and the differences from “Perceived behavioral control”.

2 - Section 1.3 should explicitly say which theory (TPB or TAM) each latent construct comes from

            A clarifying paragraph was added; see lines 296-301.

3 - When equivalence is computed across groups - is this done across non-users and users, across countries, or both?

            It is done across all groups and nationalities (Italian users, Italian non-users, Swedish users, Swedish non-users) (see lines 465-466). However, scalar invariance was computed only for the latent variables habits and climate morality). The reason is that the phrasing of these items was the same for all groups, but SN, Trust, and Control were phrased differently between users and non-users, and therefore, they were not tested for equivalence (see lines 480-484).

4 - When comparing coefficients for users and non-users in the discussion and Tables 4 and 5, is is possible to do a hypothesis test of the null hypothesis that the coefficients are the same? For most coefficients the CIs overlap, which suggests there may not be a difference, but this is an underpowered test. Furthermore, the authors also compare one coefficient (habit in italy) based on it being significant in one model and not in the other, even though the CIs overlap, which seems inconsistent.

            New χ2 difference tests were performed to measure the null hypothesis that the coefficients are the same across the four groups. Please see the lines 534-549. In this case, the CIs overlap is not a test of the difference, rather quantifying the precision of that particular parameter estimate (i.e., not the group difference). The text in the discussion section was also adjusted according to this new analysis.

5 - Similarly, the discussion notes slight differences in coefficients between Italy and Sweden. Is it possible to do a test to see if these differences are statistically significant?

            See answer to comment 4.

6 - Line 517 there is a typo - "more less"

            The typo was corrected; see line 719.

Reviewer 3 Report

Dear authors,

Your study has been very interesting to read, and the findings were intriguing too. Overall, the manuscript is well written, giving a clear background to carsharing and the motivation for the research. However, there a few things that I think can be done to make it even more interesting and easy to follow:

Introduction and theory:

1. I did not find any definition for carsharing in the introduction. A classification of carsharing, i.e., round-trip, one-way etc. is presented but I think it can be important to clarify what is referred to as carsharing and how it is distinguished from closely related concepts like car renting or carpooling that are often confused with it.
2. When listing the classifications for carsharing, a description for peer-to-peer would also be a good addition for purposes of clarity especially since P2P is still quite novel and not very widespread.
3. I found the last sentence before 1.1 that states “Any study has investigated differences between the use of carsharing services across groups…” quite confusing. I think there is need to clarify what is meant by “any study”
4. I think it is a good idea to use the theory of planned behavior to investigate carsharing since not much research has been done here yet. However, I found myself reading section 1.2 a couple of times before I could understand the reasoning behind your choice of theory. In the first paragraph, you list aspects of robustness and parsimony as justifications for this choice, which is good. However, before the text in third paragraph, some context introducing the theory of planned behavior would be good. I would suggest a more pedagogical approach to introducing it by providing an illustration (figure) showing how the different concepts affect behavior as well as the connection to TAM. I believe this would make it easier for the reader to follow before delving into the more detailed explanations of the model. Basically, what I mean is that by illustrating and giving a clear background on the theory of planned behavior, it could make it easier for the reader to understand its suitability for this research.
5. The caption under figure 1 is a bit confusing. I think it would be better to have separate text explaining what the different terms stand for after the figure rather than having it in the caption. Try to keep the captions simple and straight forward.

Materials and methods

1. The definition of carsharing provided in the survey does not exactly distinguish it from car renting. How can you be sure that some respondents did not misunderstand it?
2. There seems to be an imbalance as regards the items used to assess the variables, for example there is a big difference between 3 items for subjective norms and eight items for habit.

I think it is important to reflect on what impacts these factors could have on the results in the limitations.

Research questions

RQ1 and 3 are a yes/no questions. I would instead suggest rephrasing to “to what extent” for a more analytical response in the discussion.

Discussion

When reading the discussion, I found myself having to keep going back and forth to compare results with the RQs. I would have liked to see a more coherent presentation of the findings so that the reader does not have to go back and forth between the questions. Perhaps referring back to the RQs at the beginning of each discussion point would make it easier for the reader to follow. As it is right now, the presentation of findings for RQ3 and RQ3a are quite intertwined. I would have liked to see clear differences between the Swedish and Italian groups as well as users and non-users. And I guess this is a matter of preference, but I would also suggest having subheadings based on the RQs being addressed to help you structure the discussion.

Author Response

• I did not find any definition for carsharing in the introduction. A classification of carsharing, i.e., round-trip, one-way etc. is presented but I think it can be important to clarify what is referred to as carsharing and how it is distinguished from closely related concepts like car renting or carpooling that are often confused with it.

We included a better definition for carsharing at the beginning of the introduction; see lines 31-46.

• When listing the classifications for carsharing, a description for peer-to-peer would also be a good addition for purposes of clarity especially since P2P is still quite novel and not very widespread.

We added a description of P2P services; see lines 57-59.

• I found the last sentence before 1.1 that states “Any study has investigated differences between the use of carsharing services across groups…” quite confusing. I think there is need to clarify what is meant by “any study”

We reformulated the sentence; see lines 90-92.

• I think it is a good idea to use the theory of planned behavior to investigate carsharing since not much research has been done here yet. However, I found myself reading section 1.2 a couple of times before I could understand the reasoning behind your choice of theory. In the first paragraph, you list aspects of robustness and parsimony as justifications for this choice, which is good. However, before the text in the third paragraph, some context introducing the theory of planned behavior would be good. I would suggest a more pedagogical approach to introducing it by providing an illustration (figure) showing how the different concepts affect behavior as well as the connection to TAM. I believe this would make it easier for the reader to follow before delving into the more detailed explanations of the model. Basically, what I mean is that by illustrating and giving a clear background on the theory of planned behavior, it could make it easier for the reader to understand its suitability for this research.

We included a figure to illustrate the TPB model; see lines 167-171.

• The caption under figure 1 is a bit confusing. I think it would be better to have separate text explaining what the different terms stand for after the figure rather than having it in the caption. Try to keep the captions simple and straightforward.

We separated the text from the caption; see lines 171-172.

• The definition of carsharing provided in the survey does not exactly distinguish it from car renting. How can you be sure that some respondents did not misunderstand it?

All researchers involved in this project from different European countries agreed that this was a good definition of carsharing. Moreover, those who were not members of carsharing services were asked to what extent they were familiar with the concept. Those that were not familiar did not answer the questions of the survey about carsharing. We think that the part of the definition “No separate written agreement is required each time a member reserves and uses a vehicle” separates it from car rentals. This definition was not an issue during the pilots. Also, the participants could write comments in the survey before sending it in, and we never got any mention that this definition was not clear. Given these motives, we believe that the definition was clear enough for the participants.

• There seems to be an imbalance as regards the items used to assess the variables, for example, there is a big difference between 3 items for subjective norms and eight items for habit. I think it is important to reflect on what impacts these factors could have on the results in the limitations.

Since the inter-relatedness of the items within each latent variable (measured by Cronbach’s alpha) and the lower bound to the reliability (λ2) were under acceptable range given the literature on this topic, we do not see the specific limitation impacts they could present to the study. The choice for having fewer items by construct was due to time constraints, we did not want to have a long survey because it causes fatigue and affects the quality of the respondents’ answers.

• RQ1 and 3 are a yes/no questions. I would instead suggest rephrasing to “to what extent” for a more analytical response in the discussion.

We reformulated the RQs as suggested, see lines 307 and 312.

• When reading the discussion, I found myself having to keep going back and forth to compare results with the RQs. I would have liked to see a more coherent presentation of the findings so that the reader does not have to go back and forth between the questions. Perhaps referring back to the RQs at the beginning of each discussion point would make it easier for the reader to follow. As it is right now, the presentation of findings for RQ3 and RQ3a are quite intertwined. I would have liked to see clear differences between the Swedish and Italian groups as well as users and non-users. And I guess this is a matter of preference, but I would also suggest having subheadings based on the RQs being addressed to help you structure the discussion.

The RQs were restated in the discussion section to facilitate the comprehension of the text. The further analyses suggested by Reviewer 2 (see lines 534-549, 700-704) helped us clarify the differences between the groups, and we think that it could also help address this comment. Please, see lines 569-570; 580-581;585-588.