Violent and Nonviolent Death Tolls for the Gaza War: New Primary Evidence

2.1 Sampling and Fieldwork

The Materials and Methods file of the supplementary materials section provides more details on GMS sampling and fieldwork than this brief overview does and still more details are in Supplementary Information (SI).

The survey was preregistered, ethically approved by Royal Holloway University of London and obtained informed consent from each respondent. The Palestinian Center for Policy and Survey Research (PCPSR) collected the data from 2,000 households spread across 200 Primary Sampling Units (PSUs). Challenges to drawing a representative sample included widespread displacement and the inaccessibility of Northern Gaza, Gaza City, and Rafah due to ongoing fighting and Israeli evacuation orders. PCPSR, which has tracked population movements throughout the war, randomly selected 70, 41 and 89 PSUs in, respectively, census enumeration areas, built-up shelters, and tent gatherings – mirroring the population proportions by dwelling type when the survey was conducted. We strived to ensure representativeness by governorate of origin by earmarking some shelter and tent PSUs to cover the populations of the three inaccessible governorates. Figure 1 shows PSU locations and the governorate of origin that each PSU represents. We describe the selection procedures within PSUs in the SI section.

• Download figure
• Open in new tab

Figure 1:Sampling points by location, colour coded for governorate of origin, of respondents with further information on the sample and estimates

We wrote the questionnaire in English and PCPSR translated and then tested it in Arabic. The interview documents household members as of October 6, 2023, plus newborns added to these households during the war, and their post-October-7 vital status. Interviewers were trained to ensure accurate mortality reporting, focusing only on household members. Data were collected between December 30, 2024, and January 5, 2025, by ten two-person teams supported by four supervisors and other PCPSR staff. GPS tracking and real-time monitoring were among the quality control measures. Face-to-face interviews were conducted using tablets and mobile phones, with data instantly uploaded to a secure central server.

2.2. Sample Description

Table 1 shows the status, by gender and age, of all people recorded by the survey as household members as of October 6, 2023. Between the ages of 5 and 39, deaths among men greatly outnumber deaths among women. There are more men than women reported missing in all age categories, including children, especially for ages 18-39. Women make up just 1 out of 52 people reported imprisoned. The primary group in the sample who left Gaza are young adults.

Table 2 subdivides reported deaths into categories, by gender and age. Violent deaths greatly outnumber nonviolent deaths. Among violent deaths, there are many more males than females, especially between the ages of 18-39.

Table 3 gives the within-sample number of violent deaths along with the percentage dying violently within each governorate. Readers should interpret this table with caution because many people, after being displaced, will have died outside their governorate of origin, and we did not collect information on where people were killed.

The survey recorded post-October-6 births of 179 boys and 178 girls, of whom it was reported that two boys and two girls had died.

4.1 Interpretations and Conclusion

In this survey-based study of mortality in the Gaza war through January 5 2025, we conclude that: (a) the violent death toll of the war resulted in the deaths of around 3.6% of Gaza’s population, exceeding the official GMoH total; (b) around 56% of the violent deaths were among women, children, and the elderly, a percentage similar to that endorsed by the GMoH; and (c) violent deaths far outnumber nonviolent deaths, though there have still been at least several thousand nonviolent excess deaths. The proportion of deaths made up by women, children, and elderly supports the general perception in which non-combatants are frequent victims of the conflict.

There is striking alignment between our survey-based finding, that the comparable GMoH violent-death number is 34.7% below our central estimate, and the finding of (7), using completely different capture-recapture methods, that the GMoH tally is 39.1% below their central estimate.² Thus, the GMO and (7) combined provide a rare validation (21) finding for the two methodologies which are increasingly used to measure death tolls in war (22, 23).

Our findings are also incompatible with claims that: (a) the GMoH has “inflated the death toll” (5, 24); (b) indirect deaths could exceed violent deaths by at least a factor of four (10, 25); and (c) “it is likely that 62,413 people have died of starvation” (appendix to 11). Follow-up surveys might provide central estimates or distributions that differ substantially from the GMS. Nevertheless, this is the first population survey of mortality in Gaza which gives independent and grounded estimates that will anchor future discussions.

Our findings suggests that a high ratio of indirect to direct deaths is not inevitable in warfare, as some analysts seem to assume (10). This should not be surprising as, for example, the Kosovo Memory Book project classified just 281 out of 13,517 deaths in the Kosovo war (1998–1999) as indirect (26). Indeed, indirect deaths are usually only estimated (27) in settings with extremely poor conditions such as Darfur (3) or Tigray (28) and direct deaths have substantially outnumbered indirect ones even in the extremely impoverished environment of Yemen (26).

Despite a claim to the contrary (7), the GMS demonstrates that a ground survey in the Gaza Strip was feasible despite the challenging conditions. This success highlights the need for improved mortality surveillance in conflict zones.

The GMoH work has value that extends far beyond the insight it provides into the total number of violent deaths in the war and the number’s demographic composition. By naming individual victims one by one, the GMoH endows each person with a measure of human dignity (27). This project of memorialization has only begun and must continue for many years after the war ends. Estimates of numbers killed, the focus of the present paper, can help guide this casualty recording work. But we cannot provide each human being due recognition in death simply by estimating the number of deaths.

4.2 Sources of bias and uncertainty

Here we list and evaluate potential sources of both upward and downward bias in our estimates, beginning with upward bias.

Respondents may have overreported deaths, perhaps with the objective of collecting compensation for dead relatives or otherwise inflating the conflict’s mortality counts. Interviewers were clear, however, that respondents could not gain materially from their participation. Moreover, interviews began with a household roster before ascertaining the fate of each person listed on the roster. This procedure was intended to limit inflating death counts by forcing respondents to report deaths only of previously enumerated individuals, without post hoc addition of deaths.

Respondents may have reported deaths that are real but of relatives who were not household members, as seems to have happened in a survey of Iraq (28). We anticipated this possibility. Interviewer training stressed this and emphasized that relatives who are not household members must not be included in the household roster. Indeed, the mean household size in the sample (4.9) was below the population figure (5.5), arguing against inflation of households with non-member relatives.

We could not sample some populations. For example, we could not reach households that had left the Gaza Strip entirely (i.e., all members left). Although some people have left to receive medical treatment, or because of deaths of household members, it is plausible that leaving Gaza Strip is correlated with lower mortality, both because of better conditions in the destination place of residence, as well as because of selection (families with more means may have been more likely to leave). We note, however, that the number of leavers identified in our sample was relatively small (Table S4).

Finally, survey interview teams may have preferentially selected relatively high-mortality households. The survey protocol they were instructed to follow is designed to prevent this from happening, since we recognize that inadvertent factors may drive sympathetic human interviewers towards more affected families. The extent to which this may have happened is hard to ascertain and we have no positive evidence that it did happen. However, we address the possibility through a scenario in our sensitivity analysis (Table S6) that removes the data collected by the survey team that captured the most violent deaths.

A source of downward bias is our non-sampling of households with zero remaining live members, or no live members above the age of 15, whom we could interview. While the number of such households is unclear, this phenomenon has been documented by others (29).

We used shelter interviews exclusively to capture households from Northern Gaza, Gaza City and Rafah. Therefore, households from Deir al-Balah or Khan Younis that were displaced into shelters were outside our sampling frame. Thus, the estimates assume, implicitly, that mortality rates for displaced and nondisplaced households from these governorates were equal, although it seems likely that the former group suffered higher mortality than the latter group.

We were unable to reach households that remained entirely in Northern Gaza, Gaza City, or Rafah due to the extreme danger in these areas (30). This gap could contribute to underestimation. However, many displaced households experienced violence during departure, transit, or in their new locations. While we lack direct evidence on how these dynamics offset each other, the scale of displacement suggests that excluding those who remained is likely a minor source of bias.

It is unclear to what extent individuals classified as missing may be dead. This has implications not only for the overall mortality numbers, but also for the portion of the violent deaths made up by women, children, and elderly. Nevertheless, it is worth noting that, even if we make the extreme and implausible assumption that all missing persons are dead, then the central estimate for the percentage of women, children, and elderly among the dead decreases only to 52.2% (Table S5).

The above discussion of biases shows that, as is the case with all surveys, confidence intervals cannot capture all the important uncertainty surrounding the estimates. Moreover, we did not attempt to model uncertainty over the number of projected deaths in the absence of war, instead using a single number for baseline deaths in our excess death calculation while acknowledging that this is a further source of uncertainty. Readers should, therefore, view our confidence intervals as minimum ranges.

4.3 The Future

A fragile ceasefire began roughly 10 days after our data collection ended but collapsed with renewed Israeli airstrikes on March 18. The humanitarian situation has further deteriorated since then. Widespread destruction across the Gaza Strip continues to cripple institutions essential to human development, particularly the health care system. Undercounting of violent deaths by the GMoH is likely to persist, although not necessarily at the GMS-estimated rate. On April 7, six senior UN officials warned of a looming humanitarian catastrophe (31), and subsequent reports through May and early July indicate worsening conditions. The ratio of nonviolent to violent deaths may well have grown since the GMS data collection period. Past scholarship suggests that the long-term toll of conflict will probably accumulate over the coming years even if the acute conflict ends (35–37). This underscores the importance of continued GMoH surveillance, as well as independent surveys, to properly understand the unfolding dynamics of death. We hope that resources will be available to tackle this task.