Working With Croatian Municipal Population Data
If you have ever tried to pull coherent demographic trends out of Croatian census data for a smaller county, you know the pain. Dubrovnik-Neretva County is one of those places where the numbers look clean on paper but fall apart the moment you dig into them. The county covers everything from Dubrovnik's urban core to the isolated hinterlands of Konavle and the Pelješac peninsula, and each zone tells a completely different story. I spent about three weeks last year working through the 2021 census microdata and the quarterly NKP (nacionalni ključ prihoda) population registers for this county. What I found was worth the effort, but getting there required skipping around the official sources instead of relying on any single one. The central repository is DZS — Državni zavod za statistiku — and their census portal at popis1991.dzs.hr for older data and the newer electronic census release is useful, but it is structured for tabular downloads, not for tracking individual municipality-level shifts over time. You will want to supplement that with the DORA database for cadastral and address-level population registers, which carry more granular quarterly data than the census itself. The county has roughly 125,000 residents spread across five cities and three municipalities: Dubrovnik, Mljet, Orebić, Korčula, Lastovo, Cavtat, Metković, and Ploče. That breakdown matters because each zone follows a different demographic logic. Dubrovnik is aging and seasonal; Metković and Ploče are younger and more industrial; the islands have been losing population for decades. Treating them as a single unit smooths over the actual trends you are trying to study.
Here is the workflow I use. First, download the 2021 census age-sex tables for each sub-unit from DZS. Cross-reference them against the Q4 2024 DORA population register to identify net migration gains or losses that happened after census day. The DORA data has a quirk where registered address changes during the tourist season create phantom population spikes in Dubrovnik and Cavtat between May and September. I learned this the hard way when my preliminary migration model showed a +4,200 net influx in July 2023 that vanished by November. The workaround is to filter DORA records by permanent residence status and exclude temporary addresses flagged as summer residences. That drops the summer spike to a more realistic range of 600 to 900 net new registrations, which aligns better with short-term rental and utility connection data. For fertility and mortality trends, the DZS annual demographic reports are the standard source, but they publish at the županija level only. You have to break them down manually by matching birth and death records to municipality of occurrence using the DZS microdata request process. This takes time. I submitted a microdata request in early February and received cleaned age-specific fertility rates for the five city-municipality units by mid-March. The delay is not unusual, but if you are working on a tight deadline, you can approximate island and rural rates by applying the county-wide total fertility rate to the age structure of each sub-unit, then adjusting downward by 0.03 for every year of out-migration recorded in the DORA register since 2018. It is not precise, but it is close enough for most planning purposes. One thing that catches people off guard is how tourism distorts the dependency ratio. Dubrovnik's official working-age population looks healthy, but the seasonal worker population is largely self-employed or on short contracts that do not show up cleanly in employment registers. If you are modeling economic dependency or healthcare demand, you need to add an estimated seasonal cohort. I use a rule of thumb derived from the 2021 census secondary residence data and the Croatian Agency for Tourism statistics: multiply the count of secondary homes in a municipality by 1.8 and add that to the summer population estimate. The factor 1.8 comes from averaging household size and occupancy patterns across Dubrovnik, Cavtat, and Korčula. It works within a five percent margin for most summer months.
The bigger blind spot is internal displacement. The Croatian Civil Protection Agency publishes disaster and relocation data, but it is fragmented across municipal websites and occasionally missing for smaller settlements like those on Mljet or Lastovo. When I was mapping population pressure points near the Neretva River basin after the 2022 flooding events, I had to compile relocation numbers from four separate municipal PDF reports and one press release from the State Office for Displaced Persons and Refugees. No single dashboard exists. If you need this kind of data, I recommend setting up a weekly scrape of the relevant municipal gazettes and saving the PDFs locally before they get archived or updated. A word of caution about the census microdata itself. DZS applies statistical rounding to protect privacy, which means small cells — especially in Lastovo and Mljet — sometimes show totals that do not add up across age brackets. I once spent two days trying to reconcile an impossible discrepancy in the 65-to-69 female age group before realizing the rounding adjustment had created a negative count in one sub-table. The fix is to always verify that row and column sums match the published marginal totals before running any regression or projection model. If they do not, the rounding error will cascade through your calculations and give you biased results. For projections beyond the current data, the Population Council of Croatia (Narodne novine publishes their methodology annually) provides a cohort-component model that you can adapt locally. I have used it with DZS base populations and my own migration estimates. The model assumes current age-specific fertility and mortality rates continue, which is a reasonable short-term assumption but breaks down after about ten years if tourism-driven in-migration accelerates or declines sharply. In that scenario, the model overestimates population in aging coastal municipalities and underestimates it in the Neretva plain where younger families are still moving in. I have found that combining the council model with a simple gravity-based migration proxy — using distance to Dubrovnik airport and employment centers — improves accuracy for the ten-to-fifteen-year window.
Get the Full Details
The main tools I rely on are R with the demography and censusr packages for cleaning and projection, Python scripts for scraping municipal data, and a basic SQLite database to store quarterly DORA pulls. OpenRefice is useful for merging the DZS tables with DORA registers when the column headers do not match. The whole pipeline takes about eight hours per county update once it is set up, compared to the three days I used to spend doing it by hand. If you are starting from scratch and do not need microdata, the DZS Open Data portal has pre-aggregated tables for Dubrovnik-Neretva County that are sufficient for high-level trend analysis. The trade-off is that you lose the ability to detect sub-municipality shifts, which is where the actual interesting patterns live. For policy work or academic research, the extra effort of the full pipeline is worth it. For a quick overview, the open tables will get you most of the way there.
