The (limited and complicated) impact of the Bondi Terrorist Attack on wellbeing and political attitudes
This post examines changes in Australians’ wellbeing and political attitudes immediately before and after the Bondi terrorist attack, using nationally weighted survey data. Comparing responses collected in the days preceding the attack with those collected immediately afterwards, and controlling for key demographic differences, the analysis finds little evidence of broad deterioration in wellbeing, social cohesion, or democratic attitudes. The most consistent changes are reduced confidence in the federal government, lower satisfaction with the direction of the country, and lower evaluations of political leadership. Overall, the findings suggest that Australians’ core social and democratic attitudes show considerable short-term resilience
Just before 7pm on the evening of Sunday 14th December, during a Hanukkah celebration at Bondi Beach, two gunmen opened fire killing 15 people. One of the (alleged) gunmen was killed, and the other has been detained and charged by police. Scores of others (including police officers) were injured.
The shooting has been declared a terrorist incident, and is the most deadly on Australian soil in modern history. The types of investigations that would identify the full details and circumstances of the terrorist attack are ongoing, and a proposed NSW or Federal Royal Commission is yet to commence. However, there seems to be no doubt that the Jewish community was targeted, and that the attack is part of an escalation of anti-semitic verbal and physical violence in Australia since the Hamas attacks in October 2023 and Israel’s war in Gaza that followed.
On the 9th of December, 5 days before the shootings, data collection commenced for the December 2025 wave of the ANUpoll series of surveys,1 using the Online Research Unit’s (ORU’s) Australian Consumer Panel. These surveys are designed to assess the opinions of Australians on a range of important and topical issues, with each wave having one or a few areas of focus. The surveys also, however, include a number of repeated questions tracking wellbeing, aspects of social cohesion, political attitudes, and voting intentions/behaviour.
The first three days of data collection had a limited number of respondents, as the survey was piloted and tested. However, having identified no issues with the survey or collection, in-depth data collection commenced on the 12th of December. By the 22nd of December, 3,564 Australians had completed the survey.
538 respondents had completed the survey by the evening of the 14th of December, prior to the Bondi shootings. The remaining 3,026 respondents completed the survey on or after the 15th of December. As the figure below shows, the largest concentration of responses were between the 16th and 19th of December (2,577 respondents, or 72.3 per cent of the total).
Without any planning or any intention to do so, the coincidence of timing of the survey therefore creates an opportunity to estimate the immediate impacts of the Bondi shootings on Australia’s aggregate wellbeing and political attitudes. Identifying the degree of stability vs. volatility of the measures in the survey to a politically-motivated shock like the one Australia experienced this December is an important aspect of understanding and bolstering the country’s democratic resilience.
Setting up the data and analysis
To look at the impact of the shootings on our measures of democratic resilience, there is a bit of data work to do. First, the survey needs to be adjusted so it demographically reflects the population of interest (Australian adults). No survey is going to be completely representative of the Australian population. To adjust the estimates, I use a weighting technique known as Iterative Proportional Fitting, or raking, through a built in command in the data analysis software package STATA. The data is adjusted to have the same distribution as the Australian adult population by age, sex, education, and employment.
The next step is to check whether demographically, the post-Bondi survey sample is different to the pre-Bondi sample. There is a distinct possibility that the two sub-samples differ systematically, as those who are invited to participate in a survey complete it at different times over the survey period based on their time and other constraints. If there are differences and these aren’t adjusted for in the analysis, then there may be differences in measured resilience measures not because of the shootings, but because those who completed the survey after the shootings would have been different anyhow.
To test for differences in the the two samples, I estimated a probit model with the dependent variable whether or not a person in the combined sample completed the survey from Monday 15th December onwards (post-Bondi) or on or before Sunday 14th December.
Explanatory variables that were not statistically significant (even at the 10 per cent level of significance) were sex, location (lived in a capital city or not), state/territory (lived in NSW or not), education (did not complete Year 12, completed Year 12 or has a non-degree qualification, has a degree), whether born overseas, and whether speaks a language other than English at home. This lack of difference is heartening for robustness.
There are two variables that were statistically significant though - age group and employment. Older Australians (aged 65 years and over) were less likely to be in the post-Bondi sample (i.e. they completed the survey earlier). Those who were employed were more likely to be in the post-Bondi sample. In the analysis below, these variables are included as control variables, however I also include sex for interpretability and completeness.2
The next step is to set up the outcome variables of interest. From the survey, I chose variables that are particularly likely to be sensitive to a one-off shock like the Bondi shootings, rather than those that reflect more stable, long-term attitudes. Specifically, I look at the immediate impact of the Bondi shootings on 12 variables in total, separated into three broad areas of interest. These are listed below, with the variables marked with a * measured as 0/1 and those marked with a + measured on a scale of 0 to 10:
Democratic and national reflections
Satisfied with direction of the country*
Satisfied with democracy*
Has trust in Parliament*
Interested in politics*
Wellbeing and inter-personal cohesion
Life satisfaction+
Trust in others in Australia+
Perceived fairness of others in Australia+
Perceived helpfulness of others in Australia+
Political attitudes
Confident in the federal government*
Affective polarisation, measured as the absolute value of the difference in support for Labor and Liberal Parties+
Rating of Prime Minister Albanese+
Rating of Opposition Leader Ley+
The average values for each of the dependent variables are given in the two figures below (binary variables first, 0-10 variables in the second figure). The ‘whiskers’ around the estimates represent the 95% confidence intervals.
The final step is to set up an estimation model. For the 0/1 dependent variables, I estimate a probit model with the main explanatory (treatment) variable whether or not the person is in the post- as opposed to pre-Bondi sample. For the variables on the 0-10 scale, I first standardise so all the variables have a mean of zero and a standard deviation of one. I then estimate a linear regression model.3 Given the age and employment differences in the pre- and post-Bondi samples, I estimate the difference in outcomes between the two samples controlling for these core demographic characteristics.4 I also include a binary variable for whether or not a person is female.
The final methodological decision to make is how to measure the uncertainty around the estimates. Although the sample sizes are reasonably large (3,026 in the ‘treatment’ and 538 in the ‘control’ groups), any observed differences between the pre- and post-Bondi groups could be by chance, and the random differences in the outcome of interest or whether someone does a survey before or after a particular date.
It is not feasible to assume that we are making estimates from a simple random sample. Normal standard errors are therefore not appropriate. Instead, taking into account the non-probability nature of the sample, I estimate bootstrap standard errors using 250 replications. This is a technique where the software package takes repeated sub-samples from the overall sample, randomly dropping a proportion of the sample each time. I can then use how much the estimated coefficients vary across those repeated samples as an indication of how much the sample selection process is influencing the precision of the estimations.
Findings
With all the preparation described above, with multiple democratic resilience measures, and a reasonably large sample of treatment and control respondents, what did I find? What were the immediate impacts of the Bondi shootings on wellbeing and political attitudes?
Minimal, but not nothing.
Specifically, the two figures that follow show that from a statistical sense, there is not sufficient evidence to reject the null (baseline) hypothesis that there is no difference in outcomes between those who completed the survey pre-Bondi Shooting, and those that completed the survey post-shooting for nine of the dependent variables. The estimates in the figure capture the difference in average reported attitudes among respondents surveyed immediately before and immediately after the attack, holding constant age, sex, and employment. The treatment effects for nine of the variables are not significantly different from zero, even at the 10 per cent level of significance.
It is true that the marginal effects for most of these variables are in the direction that you would expect. Compared to immediately prior to the Bondi shootings, Australians had less trust in Parliament, were less satisfied with democracy, had less trust in other Australians, and were more interested in politics after the shootings. They rated Opposition Leader Sussan Ley a little lower as well (with a p-value of 0.125, or only just not statistically significant).
Three of the variables were in the opposite direction to what one might have predicted a priori though. Australians were slightly more likely to think other Australians were fair, or helpful, and they were slightly less polarised. And the difference in life satisfaction was tiny.
However, the observed differences for all of these nine variables were not large enough to conclude that there was an actual effect or difference, rather than the type of random variation in the data that would see across two sub-samples in any survey.
This leaves two variables that were significant at the 5 per cent level of significance, and one that is significant at the 10 per cent. Compared to those that completed the survey pre-Bondi,5 those that completed the survey after the Bondi shootings:
Were less likely to be confident in the Federal Government (p-value = 0.007)
Were less likely to be satisfied with the direction of the country (p-value = 0.087)
Rate Prime Minister Albanese lower on a scale of 0 to 10 (p-value = 0.042)
For those variables with a statistically significant ‘treatment effect’ at the 10 per cent level of significance, I break the post-Bondi sample into two groups and look at the timing of the shift in views.
Specifically, on Thursday 18th December, Prime Minister Albanese made his first substantial public statement post-shooting.6 This day arguably represents the threshold between the immediate post-Bondi period, and the period that reactions of the public are driven in part by responses from political leaders and parties.
I therefore re-estimate the model with two sub-treatment groups (1) completed the survey Monday 15th to Wednesday 17th and (2) completed the survey Thursday 18th to Monday 22nd.
For two of the dependent variables where the combined treatment groups were significantly different from the control group - confidence in the federal government and satisfaction with the direction of the country - there were no differences between the two sub-treatment groups.
This was not the case for the third variable - ratings of Prime Minister Albanese. For this dependent variable, there was no significant difference between the pre-Bondi group and sub-treatment Group 1 (those that completed the survey in the first three days after the attack). However, there is a large and statistically significant difference between those who completed the survey either before or immediately after the shootings, and the post-political group that completed the survey on or after the 18th of December (marginal effect for the standardised dependent variable of -0.132, p-value of 0.019).
Limitations
It is very rare that we can precisely measure the impact of shocks like the Bondi Shootings on national-level outcomes. They are not random in the sense of a randomised controlled trial. Rather they are unpredictable, which means surveys are unlikely to contain exactly the right question for the shock that is experienced. If the December ANUpoll was designed with the shootings in mind, for example, we would have asked about views on migration, anti-Semitism, Islamophobia, or gun control.
Nonetheless, the December ANUpoll gets as close as you are likely to get to a survey that measures the immediate impact of a terrorist shooting on aggregate wellbeing and political attitudes.
There are of course limitations in the analysis above. First, we only have three full days of pre-shooting data, alongside six full days of post-shooting data. It is likely that substantial differences in wellbeing and political attitudes take longer to emerge. However, I would argue that such longer term impacts require a different methodology, and capture *both* the impacts of the shootings (which can be treated as exogenous) *as well as* the impacts of the political and societal responses (endogenous). It is interesting that the only other data that I am aware of that is currently available on the shootings was collected from Wednesday to Saturday (17th to 20th), and explicitly focuses on perceptions of the federal government response.
A second limitation is that I have only measured so far simple, nation-wide, linear impacts. It is entirely possible that the impacts for some of the variables were negligible for all Australians, but significant for specific cohorts (for example those who live in Sydney, those with a baseline set of attitudes, and of course Jewish-Australians). It may also be that the effects are non-linear, requiring a more complex functional form. To the extent that sample sizes allow, these more complicated effects are important to interrogate.
A third limitation is that the sample size pre-Bondi is reasonably small. We don’t have data over a full week, with surveys undertaken on different days. A larger sample would have allowed for a more precise estimate.
Implications and reflections
With a terrorist attack like the one experienced in Bondi, it is natural to feel that nothing will ever be the same again. Of course, for the victims and their family and their immediate community, this is the case. And from a policy-setting, many of us hope that new measures will be introduced that balance protecting the community with preserving our way of life. But, we can sometimes wrongly conclude that the impacts on social cohesion or democratic resilience are also inevitable.
But, terrorists just don’t have that kind of power.
Australians didn’t automatically turn against each other, their Prime Minister, their Opposition Leader, or democracy. The balance of probabilities would suggest that Australians are a little less satisfied with the direction of the country, and a little less confident in the federal government. They also have more negative views towards the Prime Minister and, to a lesser extent, the Opposition Leader. But, the differences for the other variables are small relative to sampling and design uncertainty, and cannot be distinguished from random variation with confidence.
Author, historian, and journalist Julia Baird wrote on the Sunday after the shootings that
At times of great cultural shocks, of barely fathomable events like the sunlit massacre of innocent, faithful people, many things are inevitable: weaponisation, politicisation, rage, blame, defensiveness, strategic positioning, fear, the crossing of arms, the jabbing of fingers. Some of this may be deserved. Some mistakes will be made, too, in haste and in pain.
But if we are to inch towards unity, the more we pause to listen, the more sure-footed our path forward will be.
Her argument is to stop, and listen. Listen to Jewish Australians, and those who were injured or who lost loved ones.
I would also argue that we should also take the time to listen to the data. In the days just before and just after the Bondi-shootings, Australians are satisfied with the direction of the country, with democracy, and their own lives. They aren’t that polarised, and trust Parliament. Importantly, they think that other Australians can be trusted, try to be fair, and are helpful.
Australia’s democracy is resilient. It is up to all of use to keep it that way.
Ethics approval for the survey was given by the ANU Human Research Ethics Committee
Results are similar if I also include a more extensive set of demographic controls (location, education, etc.).
Though the conclusions are exactly the same if I use the ordered probit model on the original 0 to 10 scale
It is important for this type of analysis to not include control variables that are impacted by a person’s treatment status. It is not likely that in the short-term the Bondi shootings have had a positive or negative impact on employment for the survey sample. However, until Labour Force Survey data is made available by the Australian Bureau of Statistics, it is not possible to completely rule that out. General conclusions hold though if I do not include employment in the model.
No formal multiple-testing adjustment is applied. There are arguments for and against multiple adjustments, particularly when testing different constructs, rather than multiple proxy variables for the same construct.
Prime Minister Albanese, alongside the Minister for Home Affairs and Attorney-General made an additional substantial speech on Monday 22nd of December. However, this was too late to be captured by the survey