The manuscript reports the results of a survey study that evaluates the effect of frequent flood experience on responsibility attribution and perception of social norms while zooming in on a number of moderating variables in such a relationship. Overall, the explorative analysis and theoretical links contribute to the literature of adaptation behavior. However, some improvements need to be made in order for it to be publishable in a journal like NHESS.

Major comments

• The introduction frames the literature gap as there being limited knowledge on the drivers of responsibility attribution and norm perception, and suggest frequent flood experience as being a possible driver. Yet, from my understanding quite some studies have looked into the drivers of these two concept , as shown in the literature overview as well. The authors could elaborate how their study extend on previous literature to strengthen their research gap.
• Throughout the manuscript locus of control and self-efficacy seems to be used interchangeably, whereas only self-efficacy is included in the independent variables. While both are construct related to control, self-efficacy in behavioral literature refers to one’s confidence in being able to perform/implement a measure. Locus of control, on the other hand, refers to one’s belief whether they can control the outcome of their life. The authors could examine which concepts they are referring to, and adjust the manuscript accordingly.
• Regarding the questions used to elicit the variables, the author could elaborate on the development of these items/scales. Are these validated items/scales based on previous literature, or did the author develop the items themselves?
• Another concern relates to the representativeness of the sample, especially in relation to the generalizability of the results. Providing the socio-demographics and reflecting on the representativeness would strengthen the study.
• Most concerningly, the (polarization in) flood risk management and implications for disaster risk reduction section is written as if it is broadly applicable. Yet social norms seem to be strongly influenced by their local environment, limiting the more general recommendations.
• In the limitation section, the exclusion of covariates, such as the socio-demographics, and its implications is reflected upon. I am not suggesting that this should be included explicitly, but the reasoning for such methodological decisions could be reflected upon.

Minor comments

• No regression statistics are reported, such as the observations and adjusted r-squared per model iteration.
• Line 37, Köhler & Han (2024) should be Köhler and Han (2024)
• Line 73, Begg (2017) should be Begg et al. (2017)
• Line 106, highlighly, should this be highly?
• Appendix A1 seems to be missing confidence intervals for personal responsibility

General comments

The paper explores the relationship between frequent flood experiences towards norm perceptions and responsibility attribution. The argument for studying this relationship is well-motivated in the introduction (noting limited knowledge of the antecedents of responsibility attribution and norm perception), which makes a more convincing case for the paper’s contribution to the literature. It is interesting to see that the analysis reveals some evidence of polarization in responsibility attribution and also reveals important moderating variables, namely self efficacy beliefs and ingroup identification. Furthermore, the paper benefits from having a strong discussion section, which connects the empirical findings to the potential theoretical explanations. This was especially noticeable in the paper’s discussion on norms that builds on a strong literature review and theoretical link section. Generally throughout the paper, the arguments are also well-supported with relevant references.

However, the paper requires major revisions, particularly with regards to further justifying and refining the empirical methodology, which could potentially change the results and conclusions. The research questions (specifically RQ2 and RQ3) and the variables in the analysis (specifically the norm constructs and helplessness) can be better motivated through connections with existing literature. Therefore, I recommend on addressing these points to strengthen the credibility of the findings before the paper can be reconsidered for publication.

Specific comments

Major comments

With regards to motivating the research questions:

• The literature review and theoretical links section can be more focused towards motivating RQ2 and RQ3. The development building towards these research questions is lacking in the following three aspects:

1. The literature review on “perception of norms” covers topics on self-efficacy and ingroup identification, which are the chosen moderating variables in RQ2. This aspect is well-argued. However, the literature review on “attribution of responsibility” only discusses the potential influence of self-efficacy and the influence of ingroup identification is missing.
2. The literature review and theoretical link can better clarify how self-efficacy and helplessness is connected to perceived control. Or explain how previous literature often associates perceived control with self-efficacy and helplessness, motivating these two measures as moderating variables in the analysis. The concept of helplessness is also underexplored in the literature review and theoretical link sections, compared to other key variables of the empirical analysis.
3. RQ3 is largely unclear and it is important that there is further explanation. What does it mean when you explore the interaction of your dependent variables (norm and responsibility)? Interaction effects are normally for the independent variables. Furthermore, the methods (regression equation 8) and the results (figure 10 and figure 11) do not seem to answer the research question in a straightforward manner. Can the trends given from RQ1 also be used to answer RQ3?

With regards to the survey:

• It is good practice to include the complete questionnaire as supplementary information if possible. In Section 4.2, it can be clarified how the survey items were formulated: whether questions or statements are standard/validated survey items, have been used in previous survey research, or have been developed by the author themselves.
• In Table 1, there are three constructs for norms: injunctive, descriptive, and personal. It is the first time these terms appear, which turns out to be important concepts in the discussion section. It may be helpful to introduce these norm constructs in advance, such as whether these distinctions have been recommended in previous literature.
• It is also good practice to present the descriptive statistics of the sample, besides the main variables that are included in the analysis, for the reader to get a better picture of the sample such as in terms of their socio-demographics. The sample sociodemographic characteristics can also be compared with population characteristics to determine the representativeness of the sample. It should be made clear if this is not possible, mentioning limited data availability or that these information are not elicited in the survey. Furthermore, if the survey data provides socio-demographics, the regression models can be improved by including these as control variables. If it is not feasible, it could be mentioned in the mixed effects model with random intercepts whether some part of the respondent characteristics is taken into account by the individual specific intercept, u_j.
• It may be useful to provide a correlation matrix of the variables listed in Table 1.

With regards to the regression equation:

• Equation 2: The linear equation for norms lacks motivation. Does it serve as a baseline to test if any relationship exists? If so, there should also be a linear equation for responsibility attribution. Only having the linear equation for norms made it seem like it is intentionally for the purpose of showing an increasing relationship for norms (given the statistically significant result, even though the nonlinear effects with dummy variables are statistically insignificant). Do note that a linear equation with FFE as the independent variable may be less appropriate given that a value of 3 indicates experiencing 3 or more floods. FFE is not a continuous variable.
• The terms “n_ij” and “r_ij” used in the equations may need to be “n_i” and r_i” instead because these norm and responsibility constructs are categorical/dummy variables that applies to all individuals and do not vary for each individual j.
• Potentially crucial to consider: it may be inappropriate to include interaction effects in regression without including the main effects. This applies to equation 3-7. For instance, equation 3 should include n_i (given the correction above) so that each individual j has also different intercepts for the different norm constructs, such that the coefficients of n_i captures the average z_norm scores when FFE=0. The main effect of each FFE may also need to be included. This will make the interaction effect capture the additional differences in marginal effects on top of the main effect, instead of capturing both the main effect and interaction effect when the regression equations are as is currently written. I am also open to further discussing this point.
• The regression equation for RQ3 remains unclear. What do each of the nr_ij represent? Is it written with the appropriate notation? It seems that equation 8 estimates equation 3 and 4 jointly, instead of separately. It is unclear how this equation answers the research question RQ3. Looking at the corresponding results in Figure 10 and 11, what is the reason for choosing comparing injunctive norm with own responsibility and descriptive norms with collective responsibility, respectively? Testing these comparisons is not clear given regression equation 8 and RQ3.

With regards to the limitations and future research:

• It could be raised as a limitation that the study treats all previous flood experience equally, not taking account the variations in severity or impact of previous flood experiences. The paper can further reflect upon this limitation, while it is only briefly mentioned in future research section, line 512-513. For example, it could be the case that the nonlinear effects found after the first flood experience are mainly driven by a severe flood, such as the 2002 flood mentioned in line 181-183.
• An interesting direction for future research that can be mentioned in the paper is to survey the theoretical mechanisms that the paper has identified or hinted, but were not explicitly tested. For example, do frequent flood experiences increase social interactions with different actors in FRM (as in line 139-141)? Or does frequent flood experiences change the evaluation of the effectiveness of private protective behaviour (as in line 143-144)?

Minor comments

• In the introduction: Perhaps in line 50 before the sentence, “In Section 2…”, also mention that the paper will explore moderating variables (self-efficacy beliefs and ingroup identification), since it is also an insightful finding of the paper.
• Line 203: What was meant with iteratively adjusted the reference group? It should not matter which one is the reference group. The estimated coefficients for the flood experience dummies should all be relative to the reference group. So, there should not be a need to specify that the regression is estimated separately with different reference group if this is what was meant by iteratively adjusting the reference group.
• Line 245-246: It is also common to test the coefficients against a significance level of 0.01, 0.05, and also 0.1. So, it does not have to be elaborated that p-values that slightly exceed 0.05 will still be interpreted as statistically significant. The coefficients that are significant at a 0.1 level can be interpreted as having “weak evidence”, “some evidence”, or “marginally significant” for the direction of the relationship.
• Showing the results visually with a line graph is practical. Perhaps, what can be considered is adding the standard errors to the points in the line graph so the readers can more easily infer the statistical significance.
• Additional regression statistics can be shown in Table 3 and Table 4, such as sample size, f-statistic, and adjusted r-squared. This also applies to the tables in the appendix.

Technical comments

• Abstract line 16-17: Specify how the gap in responsibility attribution widens. That responsibility attributed to the city/state increases, while responsibility attributed to self/the community increases.
• Abstract line 17-18: Also specify in what way norms are less dynamic. That results show that norms perception weakly increases with frequent flood experience.
• Line 220: Do not write “n_ij/r_ij” in the text to prevent confusion. It could be initially thought as a fraction. Instead what can be written is “To compare the effects of flood experience between different norm and responsibility constructs, we added n_ij and r_ij as interaction effects, respectively.”
• Line 229: Can replace “low self efficacy” with just “self efficacy” for consistency, since the moderating variables are categories of perceived helplessness and self-efficacy.
• Equation 6: the notation of the dependent variable can be consistent with how the dependent variable is written in other notations.
• Tables in the appendix: Instead of writing the p-values that are just above 0.05 in the cells, you can give a note below the tables indicating three significance levels, “*: p<0.1; **: p<0.05; ***: p<0.01”.