Dear Reviewer #1,

We appreciate the time and effort you dedicated to provide feedback on our manuscript and are grateful for the insightful comments and valuable improvements to our paper. We diligently went through your remarks and corrected our manuscript accordingly.

Please find our responses and revised text sections (in italics) below each of your comments.

1. Theoretical Framework and Indicator Selection:

• The paper lacks a clear theoretical foundation for the selection of adaptive capacity indicators. Please provide a comprehensive literature review that explains the theoretical basis for using these indicators. Perhaps authors can justify why you chose to focus on indicators rather than established theories of adaptive behaviour.

Thank you for pointing out this important issue. To clarify the aim of the study and the rationale behind our selection of indicators, we will include the following paragraph in line 82:

We conducted a review of the current literature to identify the most commonly used adaptive capacity indicators, irrespective of our own judgement on whether or not these indicators explain adaptive actions. While many studies focus on explaining household adaptive behaviour (e.g., Bamberg et al., 2017; van Valkengoed & Steg, 2019), we restricted our review to papers that explicitly address the concept of adaptive capacity. We scanned the current literature from the Web of Science and Scopus databases, searching for peer-reviewed articles with “adaptive capacity” in the title. From this body of literature, we identified a) highly-cited conceptual papers and reviews on adaptive capacity indicators at the household-level (e.g., Whitney et al., 2017, Mortreux and Barnett, 2017; Cinner et al., 2018; Siders, 2019), and b) quantitative empirical papers on the capacity-action relationship (e.g.; Grothmann and Patt, 2005; Mortreux et al., 2020; Barnes et al., 2020; Green et al., 2021; Bartelet et al. 2023). This procedure aligns with our aim to test the usefulness of commonly employed adaptive capacity indicators as proxy for adaptive behaviour. Appendix X offers an overview of identified references for indicator selection, their research question, context, theoretical framework, and main findings.

We compiled a comprehensive list of indicators from the studies, regardless of conflicting findings or null results, and grouped similar ones together. Some indicators were excluded as they were specifically referring to resource-dependent communities. For example, while livelihood diversification is often understood as a form of flexibility in societies with a natural resource-based economy, we considered this capacity not relevant in our study setting. This process resulted in 18 indicators representing adaptive capacity of households in the German pluvial flood context. Although we derived the indicators empirically, many of them are also grounded in the theoretical frameworks mentioned above (e.g. protection motivation theory, sustainable livelihoods framework). Table 1 presents an overview of the indicators, their operational definitions, key references, and theoretical foundation.

• Add a detailed section explaining how each indicator (both generic and specific) was selected, defined, and operationalized. This should include: a) The rationale for including each indicator b) How each indicator is measured (e.g., survey questions, scales used)

Thank you for pointing this out. Some of the information you mention is already presented in Tables 1 and 2. Unfortunately, both tables are not correctly positioned in the current template. We will ensure that both tables are correctly positioned within the text (and not after the references) in the final manuscript.

To address your helpful comment, we have made the following revisions:

a) We have added a paragraph to the main text body that outlines the general rationale for indicator selection (see also comment above).

While many adaptive capacity indicators can be linked to established theoretical frameworks, we found surprisingly few empirical studies explicitly reference these theories. To clarify that several of the indicators in our study are indeed grounded in theoretical frameworks, we have added an additional column to Table 1, indicating relevant theoretical frameworks for each indicator (see comment above and the paragraph we will add in line 82).

b) The scales for each indicator are already presented in Table 2. To clarify that our questionnaire (containing both questions and answer options) is available as an open-access resource, we have added a sentence in the method section (line 124):

“This process resulted in a questionnaire with an average length of 36 minutes which covered a broad range of topics such as perceptions about climate change and extreme weather events, risk awareness, pluvial flood damage and event characteristics, private flood risk adaptation measures, housing characteristics, and sociodemographic characteristics. The questionnaire is openly available (Schubert et al. 2024).

• Provide a more comprehensive review of previous findings on both generic and specific indicators. This could include: a) A summary table of key studies, their indicators, and main findings. b) Discussion of any conflicting results in the literature and how your study addresses these conflicts.

Thank you for your suggestion to strengthen our manuscript by making more explicit references to the literature. In response to your feedback, we have made the following revisions:

a) We have added a table in the Appendix that provides an overview of the studies, summarizing their research question, context, theoretical framework, indicators and main findings (see also our first comment).

Additionally, we have revised the discussion section, where we now explicitly reference these key studies and discuss our findings in relation to them. We amended the following lines on page 26:

Line 365:

Owning a property as well as having a larger social network makes flood

risk adaptation more likely; both effects are also well documented in the adaptation literature (for ownership, see Grothmann and Reusswig 2006, Kuhlicke et al. 2020, Dillenardt et al. 2022; for social network, see, for example, Adger 2003, Pelling and High 2005). Similar positive effects for social capital have also been reported in the capacity-action literature (Barnes et al. 2020, Bartelet et al. 2023).

Line 368:

The finding that neither wealth nor income are drivers of adaptation action at the household level is consistent with studies on household flood adaptation in Germany (Grothmann and Reusswig 2006, Dillenardt et al. 2022), as well as previous findings on the capacity-action relationship (Mortreux et al. 2020, Barnes et al. 2020, Green et al. 2021).

Line 378:

The importance of these factors has also been demonstrated in recent meta-analyses (Bamberg et al., 2017; van Valkengoed and Steg, 2019), various flood-related studies (e.g. Grothmann and Reusswig, 2006; Bubeck et al., 2023; Dillenardt and Thieken, 2024) and with the capacity-action literature (Mortreux et al. 2020, Barnes et al. 2020, Bartelet et al. 2023).

b) Thank you for this suggestion. We think that the available studies on the capacity-action relationship – the central focus of our analysis – currently hardly allow for a comparison. We included the following paragraph in line 81 to make that more transparent:

Overall, the capacity-action research field is still emerging, and strategic meta-studies are lacking. So far, only a small number of studies with a set of diverse indicators conducted in very different study contexts exist. Thus, findings are difficult to compare across case studies.

We additionally want to point out that we included all indicators in our review, irrespective of conflicting findings or null results (see first comment, which will be included in line 82):

We compiled a comprehensive list of indicators from the studies, regardless of conflicting findings or null results, and grouped similar ones together.

1. Methodology:

• Regarding the missing data: The assumption that income data is Missing At Random (MAR) is a crucial one that requires careful consideration. Income data often has patterns of missingness that may be related to the income levels themselves, potentially making it Missing Not At Random (MNAR). I suggest followings: a) Describing any tests or analyses performed to investigate the missing data mechanism, b) Discussing the plausibility of the MAR assumption for income data in your specific context. c) If you suspect the data might be MNAR, consider discussing potential implications for your analysis and results.

We fully agree with you that this point was not presented explicitly enough in our manuscript.

We explored the missing data patterns and mechanisms with graphical diagnostics provided by the VIM package (Templ et al. 2012). More specifically, we used matrix plots, margin plots and mosaic plots to detect relationships between the values of different variables and the propensity to be missing (see Rcode 04-imputation-missing-data.R in the assets section).

However, "it is not possible to test MAR versus MNAR since the information that is needed for such a test is missing" (van Buuren 2018, p. 36). To make the MAR assumption more plausible, we estimated a predictor matrix and included all correlated variables as predictors (van Buuren 2018, p.167). E.g. for the income variable, 33 variables were used to predict missing values, amongst them variables such as age, gender, education and living area.

While we can explain parts of the missingness with the imputation models, you are indeed right that is very likely that another part remains unexplained (MNAR). However, this is not problematic for two reasons. Firstly, a simulation study has demonstrated that multiple imputation is remarkably robust against MNAR (Collins et al. 2001). Secondly, even when falsely assuming MAR, results are still less biased than a complete case analysis, which would only be unbiased under MCAR (van Ginkel et al. 2020).

We decided to add a short paragraph about the MAR assumption to our revised manuscript, thereby discussing the assumption in general and not only with reference to the income variable. The following sentences will be included in line 159:

Missing data patterns and mechanisms were explored with graphical diagnostics from the VIM package (Templ et al. 2012). Multiple imputation generally starts from assuming a missing at random (MAR) mechanism (van Buuren 2018, p. 165). To make this assumption more plausible, we estimated a predictor matrix and included all correlated variables as predictors (van Buuren 2018, p.182). Since distinguishing between MAR and missing not at random (MNAR) is not possible (van Buuren 2018, p. 36), we cannot rule out the presence of MNAR in our data. Nevertheless, multiple imputation is remarkably robust against MNAR (Collins et al. 2001), and even if MAR is falsely assumed, estimates remain less biased than those from a complete case analysis (van Buuren 2018, p.57).

• To demonstrate the robustness of your findings, I suggest presenting a sensitivity analysis such as with and without imputed data (complete case). This could go into Appendix.

We are happy to provide the results of the complete case analysis in the Appendix. Overall, the results are similar; however, the models using imputed data show more pronounced effects, with stronger effect sizes and smaller p-values.

Nevertheless, we prefer not to frame this as a sensitivity analysis for two reasons. First, the complete case (CC) regression models suffer from a loss of statistical power. This is particularly evident in models M3 and M6 (tenant models), where the sample size is too small given the large number of predictors. Second, the complete case results are likely biased due to the violation of the MCAR assumption (see also the previous comment). Therefore, comparing the imputed and complete case results does not provide meaningful insights into the robustness of our findings.

We have added the table in the Appendix and the following paragraph in line 162:

We also analysed the subset of complete cases and obtained similar findings (see Appendix B2). A comparison of the p-values and effect sizes reveals that the multiple imputed models (Appendix B) are more efficient than a complete case analysis.

1. Results and Discussion:

• In the discussion section, elaborating more on the practical implications of your findings for policy makers and flood risk managers could enhance the section.

Thank you for this suggestion, which was also brought up by the second reviewer.

We have added another paragraph on policy implications in the discussion section of the revised manuscript (line 431):

5.4 Policy implications

Based on our findings, we recommend two key policy measures to enhance local adaptive capacity and household adaptation efforts: a) promoting local adaptation information and participation initiatives (e.g., led by municipalities) to strengthen risk awareness and self-efficacy among citizens, and b) creating targeted funding programs or financial incentives aimed at supporting low-income households.

Our results demonstrate that measures which increase specific capacity are key and benefit all societal groups. Risk perception and previous risk experience are the strongest drivers of adaptation actions for both homeowners and tenants. Unlike generic capacity, specific capacity, such as risk awareness, “can potentially be altered within the short to medium term, and the power to do so lies at least partially with local policy makers” (Werg et al., 2013, 1614). Municipalities could play a key role in this, for example by hosting information events to inform citizens or by sharing experiences of affected residents and successful adaptation efforts. However, recent surveys and research show that the majority of German municipalities are still not actively informing citizens about flood risks and protection measures (von Streit et al., 2024; Friedrich et al. 2024), let alone engaging them in risk management (Wamsler, 2016).

Another major finding of our study is that income groups in our sample differ in how they translate their financial assets into adaptation actions. This suggests that broad ‘scattergun approaches’ like tax incentives or public funding may be less effective than differentiated measures and interventions targeting underprivileged groups. While medium- and high-income households have the financial capacity to implement adaptation measures, they often fail to fully realise this potential due to a lack of specific capacity. For these groups, policy should focus on enhancing risk awareness, self-efficacy, and motivation for protective action, whereas funding programmes are crucial for low-income households to enable the implementation of more costly adaptation measures.

1. Presentation Improvements:

• At the beginning of each analysis section, add a brief paragraph or sentences stating the purpose of the analysis and how it relates to your overall research questions.

We acknowledge that additional guidance would help readers navigate the results section. Therefore, we have added the following introductory paragraph to Section 4 Results (line 182):

To explore whether adaptive capacity translates into adaptation, we first take stock of the households’ adaptive capacity and adaptation actions in our sample using descriptive statistics. Subsequently, we utilise correlation and regression analysis to examine how adaptive capacity influences households' decisions to implement pluvial flood adaptation measures.

• Consider renaming "specific capacity" to a more descriptive term.

While we appreciate the reviewer’s feedback, we have respectfully decided not to universally rename "specific capacity." The term was introduced in a highly cited paper by Eakin et al. (2014) and is widely recognized in the field, as well as utilised in IPCC assessment reports (e.g., Castellanos et al. 2022, p. 1748).

However, we acknowledge the importance of being precise regarding the specific hazard examined. Therefore, we have replaced the term "specific capacity" with "flood-specific capacity" in the captions of Figures 6 and 7 and in Appendix B.

• Average duration of residence cannot be fully translated as place attachment. I would state it as “average duration of residence” only.

Thank you for pointing this out. We have changed the term in the text body and all figures and tables to “duration of residence”.

Technical Corrections:

• Please specify the software used for statistical analyses.

We have added the following sentence in line 181:

All analyses were performed with the statistical software R (Version 4.3.1).

• Line 140: Please include the actual response rate figure in the body text.

We have added the response rate in line 140 as follows:

Despite efforts to increase the response rates such as a mixed-mode design, response rates were rather low (8 % for the randomly selected households and 5 % for the purposive sample).

We hope that the revisions and responses we have provided adequately address your comments. We sincerely appreciate your thorough review and the valuable feedback that has helped enhance our manuscript.

Kind regards on behalf of all authors,

Annika Schubert

Literature

Castellanos, E., M.F. Lemos, L. Astigarraga, N. Chacón, N. Cuvi, C. Huggel, L. Miranda, M. Moncassim Vale, J.P. Ometto, P.L. Peri, J.C. Postigo, L. Ramajo, L. Roco, and M. Rusticucci (2022): Central and South America. In: Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [H.-O. Pörtner, D.C. Roberts, M. Tignor, E.S. Poloczanska, K. Mintenbeck, A. Alegría, M. Craig, S. Langsdorf, S. Löschke, V. Möller, A. Okem, B. Rama (eds.)]. Cambridge University Press, Cambridge, UK and New York, NY, USA, pp. 1689–1816, doi:10.1017/9781009325844.014.

Collins, L.; Schafer J. & Kam C. (2001): A comparison of inclusive and restrictive strategies in modern missing data procedures. In: Psychological Methods 6 (4): 330–351. https://doi.org/10.1037/1082-989X.6.4.330

Friedrich, T.; Stieß, I.; Sunderer, G.; Böhmer, C.; Murawski, W.; Knirsch, F.; Otto, A.; Wutzler, B. & Thieken, A. (2024): Kommunalbefragung Klimaanpassung 2023. Climate Change 34. Online: https://www.umweltbundesamt.de/publikationen/kommunalbefragung-klimaanpassung-2023 [4.10.2024].

Templ, M.; Alfons, A. & Filzmoser, P. (2012): Exploring incomplete data using visualization techniques. In: Advances in Data Analysis and Classification 6, 29–47. https://doi.org/10.1007/s11634-011-0102-y

van Buuren, S. (2018): Flexible imputation of missing data. 2^nd edition. New York: Chapman and Hall/CRC.

van Ginkel, J. R.; Linting, M.; Rippe, R. C. A. & van der Voort, A. (2019): Rebutting Existing Misconceptions About Multiple Imputation as a Method for Handling Missing Data. In: Journal of Personality Assessment 102 (3), 297–308. https://doi.org/10.1080/00223891.2018.1530680

Dear Reviewer #1,

We greatly appreciate your insightful comments regarding the theoretical framework and indicator selection in our study. Following further discussion within the author team, we decided to undertake more substantial revisions in Chapter 2 than those initially outlined in our response from October.

In summary, we

• Renamed the title from “Conceptualising adaptive capacity” to “Conceptualising and measuring adaptive capacity” to emphasise that the chapter also discusses the indicator selection
• Reorganized the chapter and divided it into two distinct subsections:

1. “2.1 Evolution of the concept” outlines the evolution of the concept and how its understanding has changed over time. This was already part of the first draft.
2. “2.2 Identifying adaptive capacity indicators” was newly developed in response to the reviewer's helpful suggestions. It provides a detailed account of our literature-based indicator selection process. Additionally, we have included a new table in the Appendix (A1), which summarises the literature used for our conceptualization and indicator selection. Furthermore, Table 1 has been expanded with an additional column to indicate the relevant theoretical frameworks for each indicator.

We have also updated our responses in the Author's Response PDF to reflect these changes.

Kind regards on behalf of all authors,

Annika Schubert

Dear Mr. Momin,

Thank you very much for taking the time to review our manuscript. We are grateful for your positive feedback regarding our work and appreciate the specific issues you raised and the valuable questions you have posed.

Please find our responses, along with the revised text sections (in italics), provided below for each of your questions.

1. Question: If 1,571 responses are considered a low response rate, what number of responses would be considered desirable?

Thank you for pointing out that our sentence in line 140 about the low response rate is misleading. The overall number of responses (n = 1,571) is sufficient to ensure statistical power; however, we had to invite many households to achieve this number. In line 140, we refer to the response rates indicated in Figure 3. To make that more clear, we add the response rates in line 140:

Despite efforts to increase the response rates such as a mixed-mode design, response rates were rather low (8 % for the randomly selected households and 5 % for the purposive sample).

1. Question: Given that most households were educated, wealthy, and informed, why did the online approach (i.e., link shared in local newspapers and Facebook advertisements) not perform well?

Thank you for this very interesting question. Unfortunately, it is not possible to calculate an exact response rate for the publicly available questionnaire (convenience sample), as we do not have data on the number of households that saw the ads. Even though respondents from the convenience sample could also choose between an online questionnaire or a telephone interview, most respondents opted for the online version. While we do have some metadata from the Facebook ad (e.g. impressions, clicks), it remains speculative why this innovative method yielded rather disappointing results. One possible explanation could be the generic and non-personalized nature of the advertisement.

1. Question: Could the authors elaborate on potential improvements in data collection methods or strategies to convert non-responses into responses, aiming for a response rate exceeding 50%?

Survey response rates have been declining significantly over time, and even large-scale survey programs that employ high-quality methodologies nowadays rarely achieve response rates exceeding 50% (see, for example, ESS 2024). Numerous studies have explored strategies to increase response rates, including offering different response modes (such as mixed-mode surveys), optimizing questionnaire length, carefully crafting the wording of invitation letters, providing incentives, and adjusting the timing and frequency of reminders (e.g. Groves et al 2009, p. 201f.).

However, the implementation of these strategies is highly dependent on available resources, particularly time and budget constraints. In our household survey, we employed several of the already outlined strategies and mentioned them in the manuscript, including a mixed-mode design and pretests to refine the questionnaire (see lines 128 & 140).

1. Question 4: Shouldn’t the surveys be targeted to reach more low-earning (or low-risk taking) respondents more effectively

Indeed, increasing the number of low-earning respondents should be an important aim of future surveys to test exactly the hypothesis you mentioned. Including “hard-to-reach” subgroups such as those living in vulnerable social and/or economic situations in survey research is often difficult due to sampling issues and individual barriers to participation (Ellard-Gray et al. 2015). As exploring income effects was not the primary objective of our study, we did not establish a quota for high-/low-income households.

Even though low-income households were generally underrepresented in our survey (see Appendix 1), our data-driven classification of low- and high-earning households based on quantiles (15% and 90%) allowed us to estimate coefficients for the subgroups.

For clarification, we added the following two sentences in line 331:

Additionally, we account for differences between income groups. Households with an equalised disposable net income below 1,300 € (10% quantile) were classified as low-income, between 1,300 € and 4,000 € as middle-income and above 4,000 € (85% quantile) as high-income. These data-based income groups are roughly in line with official classifications for Bavaria (vbw 2023, p. 37).

1. Question: Could the authors provide further insights on why affluence alone is not enough for effective adaptation and what can improve practical adaptation strategies?

Thank you very much for pointing out that we could strengthen our manuscript by elaborating more on the practical relevance of our findings. A similar comment was also brought up by the other reviewer.

We have added another paragraph on policy implications in the discussion section (line 431), also taking into account the thoughts you provided in your comment on Policy implications.

5.4 Policy implications

Based on our findings, we recommend two key policy measures to enhance local adaptive capacity and household adaptation efforts: a) promoting local adaptation information and participation initiatives (e.g., led by municipalities) to strengthen risk awareness and self-efficacy among citizens, and b) creating targeted funding programs or financial incentives aimed at supporting low-income households.

Our results demonstrate that measures which increase specific capacity are key and benefit all societal groups. Risk perception and previous risk experience are the strongest drivers of adaptation actions for both homeowners and tenants. Unlike generic capacity, specific capacity, such as risk awareness, “can potentially be altered within the short to medium term, and the power to do so lies at least partially with local policy makers” (Werg et al., 2013, 1614). Municipalities could play a key role in this, for example by hosting information events to inform citizens or by sharing experiences of affected residents and successful adaptation efforts. However, recent surveys and research show that the majority of German municipalities are still not actively informing citizens about flood risks and protection measures (von Streit et al., 2024; Friedrich et al. 2024), let alone engaging them in risk management (Wamsler, 2016).

Another major finding of our study is that income groups in our sample differ in how they translate their financial assets into adaptation actions. This suggests that broad ‘scattergun approaches’ like tax incentives or public funding may be less effective than differentiated measures and interventions targeting underprivileged groups. While medium- and high-income households have the financial capacity to implement adaptation measures, they often fail to fully realise this potential due to a lack of specific capacity. For these groups, policy should focus on enhancing risk awareness, self-efficacy, and motivation for protective action, whereas funding programmes are crucial for low-income households to enable the implementation of more costly adaptation measures.

1. Question: Is there a better way to phrase this statement without using the word “hint”?

We replaced the word “hint” with indicate (line 225):

Similar to Rözer et al. (2016), our results indicate that information is more frequently obtained by those households who already experienced a pluvial flooding event.

1. Question: What does "Complete cases" refer to? Clarifying this term helps readers understand the completeness and reliability of the dataset.

Thank you for pointing out this issue. We clarified the term by adding the following explanation in line 162:

By this means, the sample size increased from 1,020 complete cases (without missing data on the variables of interest) to 1,571 households.

We hope that with the comments and changes provided, we sufficiently answered your questions. Thank you again for your thoughtful review and for helping us to improve our manuscript.

Kind regards on behalf of all authors,

Annika Schubert

Literature

Groves, R.; Fowler, F.; Couper, M.; Lepkowski, J.; Singer, E. & Tourangeau, R. (2009): Survey Methodology, 2nd Edition. Chapter 6: Nonresponse in sample surveys. Hoboken: Wiley, p. 183-211.

Ellard-Gray, A.; Jeffrey, N.; Choubak, M. & Crann, S. (2015): Finding the Hidden Participant: Solutions for Recruiting Hidden, Hard-to-Reach, and Vulnerable Populations. International Journal of Qualitative Methods, 14 (5). https://doi.org/10.1177/1609406915621420

ESS - European Social Survey (2024): Modes of Data Collection. Online: www.europeansocialsurvey.org/methodology/methodological-research/modes-data-collection [04.10.2024].

Friedrich, T.; Stieß, I.; Sunderer, G.; Böhmer, C.; Murawski, W.; Knirsch, F.; Otto, A.; Wutzler, B. & Thieken, A. (2024): Kommunalbefragung Klimaanpassung 2023. Climate Change 34. Online: https://www.umweltbundesamt.de/publikationen/kommunalbefragung-klimaanpassung-2023 [4.10.2024].

vbw - Vereinigung der Bayerischen Wirtschaft e. V. (2023): Faktencheck Verteilung – Bayern und Gesamtdeutschland im Vergleich. München.