How I Analyze Obituaries For Recurring Patterns Across Small-Town Local Papers
I spent three months importing every obituary from the Daily Advance into a spreadsheet. The paper covers Pittsylvania County, Virginia and the surrounding rural area. I needed to understand what actually gets emphasized when a town publishes death notices at scale. Here is what I found, the method I used, and where it breaks down. The process starts with raw text extraction. I scrape or manually copy each obituary into a CSV file along with the date of publication, the deceased person's age at death, gender, and the names of surviving family members listed. Then I run a basic natural language processing pipeline using Python with spaCy. The pipeline does four things: it extracts all person names to build family networks, it tags religious references, it flags military service mentions, it identifies occupations and institutions like churches or volunteer organizations, and it runs sentiment-adjacent keyword frequency analysis on descriptive phrases. The first thing you will notice if you actually look at the output instead of assuming what the data says is that religious language dominates. Roughly 78 percent of the obituaries I processed included phrases like "called home," "rest in peace," "eternal rest," or explicit references to specific churches. This is not surprising if you have ever read a single rural Virginia death notice. What is interesting is that the denomination matters less than you would expect. Baptist and Methodist phrases appear at nearly identical rates, and the paper generally uses broad Protestant phrasing rather than specific liturgical language. The pattern holds across decades of archived notices going back to the early 2000s.
The second major theme is occupational identity. I found that approximately 62 percent of obituaries mention the deceased's job, trade, or career in at least one sentence. The most frequently named occupations are farmer, factory worker, truck driver, and nurse. A counter-intuitive finding here is that the obituaries for younger deceased individuals actually list occupations more often than those for elderly people. This seems to happen because younger deaths tend to be framed as lost potential, so the writer emphasizes what the person did. Older deaths are more likely to be framed through family roles alone. I spent several weeks trying to confirm this trend with chi-square testing before I accepted that it was real and not just an artifact of how many elderly obituaries simply say "survived by his wife of forty-seven years" with no other biographical detail. Military service is the third reliable theme. About 34 percent of the obituaries reference some form of military affiliation. The overwhelming majority are Vietnam era service members. Korean War and World War II mentions appear but at much lower frequencies consistent with the demographic. What many analysts miss is that the phrasing around military service shifts over time. Notices from the 1990s tend to say "served his country" without specifying branch or rank. Notices from the 2010s and later are much more likely to include specific details like "2nd Lieutenant, U.S. Army" or "Vietnam veteran, received Bronze Star." This suggests both a cultural shift toward emphasizing individual achievement and a demographic shift where surviving veterans are more likely to be younger or whose families are more inclined to research and include those details. Farm and agricultural ties show up in ways that are easy to overlook if you only count explicit mentions of "farmer." I developed a coding scheme where any reference to livestock, crops, land inheritance, county extension services, FFA, 4-H, or farming equipment counts as an agricultural theme. Under this broader definition, roughly 41 percent of obituaries contain at least one agricultural reference. This makes sense for the region but the nuance is important. Many people listed as retired teachers or clerks also had farming families or inherited land. The agricultural identity persists even when it is not the primary occupation listed. I noticed this because my initial keyword search for "farm" was returning too few results until I expanded the codebook to include related terms.
The family structure theme deserves its own attention. The vast majority of obituaries follow a predictable survival pattern: spouse, children, grandchildren, and sometimes great-grandchildren. What is worth analyzing is what gets left out. Parent names are rarely mentioned for adults over fifty. Siblings appear in about 44 percent of notices but usually as a brief list without relationship context. Ex-spouses are almost never mentioned. Second marriages create interesting anomalies in the data. I ran into a case where a man's obituary listed a current wife and five children from her but omitted two adult children from a previous marriage entirely. The exclusion was notable enough that mutual acquaintances commented on it locally. This kind of omission is a real feature of the genre and it signals family dynamics that the published text conceals rather than reveals. You have to read between the lines the same way a genealogist would. One practical problem I encountered during this work involves name disambiguation. The Daily Advance serves a population where naming conventions create significant duplication. Three men named James Robert Wilson died within a eighteen-month period. All three were listed as retired from the same local manufacturing plant. My initial analysis clustered them together and produced garbage thematic output until I cross-referenced ages, surviving family names, and exact street addresses from public records. The workaround was building a matching key from at least three independent identifiers before running any aggregation. I now require age, spouse name, and either occupation or residence location to treat two records as the same person. This added about four hours of manual verification to the project but prevented what would have been a fundamentally flawed dataset. If you want to replicate this, the tooling is straightforward. I used Python with the following stack: pandas for data handling, spaCy for named entity recognition and linguistic processing, NLTK for frequency analysis, and a custom dictionary of region-specific keywords I built by reading through two hundred obituaries by hand. The dictionary approach is critical because off-the-shelf NLP models will miss local references like specific church names, county landmarks, and regional occupational terms. You cannot skip the manual dictionary construction step. It takes about a day of focused reading to build a usable one for any given local paper.
Get the Full Details
There are real limitations to this kind of analysis and I want to be direct about them. Obituaries are written by living relatives who have strong incentives to present the deceased in a favorable light. The themes you extract are not objective cultural truths. They are curated self-presentations shaped by family grief, social expectations, and the newspaper's editorial standards. The Daily Advance, like most small-town papers, has an unstated but very real expectation that obituaries will conform to certain norms. Deviations are rare and noticeable. The analysis shows you what the community considers appropriate to publish about death, not necessarily what people actually believed or experienced. Treat the data as evidence of social performance rather than biographical accuracy. Another limitation is selection bias. People who cannot afford to pay for a full obituary do not appear in this dataset. The Daily Advance charges by line or by word count. This means the sample skews toward middle-class and above households. Lower-income deaths may appear only in brief classified notices or not at all in the paper. Any thematic conclusion drawn from this corpus implicitly excludes the population most likely to die young from preventable causes and least likely to have family resources to write a detailed notice. Be careful about generalizing beyond the visible sample. A third issue is temporal drift. Writing conventions change. The number of words per obituary has increased steadily over the past twenty years. This means thematic density is not constant across time periods. A 2003 obituary averaging one hundred twenty words will naturally contain fewer themes than a 2023 obituary averaging two hundred eighty words. If you are comparing themes across decades, you need to control for length or you will mistake verbosity for thematic richness. I solved this by calculating theme frequency per one hundred words rather than as raw counts.
The most useful outputs from this kind of analysis are community health approximations and genealogical mapping. The occupational and agricultural data gives you a historical record of local industry shifts that official statistics sometimes miss. The family network extraction can reveal migration patterns when children consistently appear listed as residing in distant states. The religious affiliation data tracks denominational change over time. None of this is groundbreaking on its own. Combined, it gives you a grounded picture of how a specific rural community processes death publicly. If you are starting this work yourself, begin with a small controlled sample before scaling up. Pull three months of daily obituaries and code them by hand first. You will learn more from ten hours of manual reading than from running an automated pipeline on five thousand records and getting confusing results. The manual step teaches you what vocabulary to include in your dictionary and which themes are actually frequent versus which ones are noise. Once you have a working codebook, automate the extraction and spend your time on verification rather than discovery.
