Doing Research On Crime And Justice

Where to actually start when you are researching crime and justice topics

Most people waste weeks chasing the wrong sources. I spent three years pulling data from juvenile court records and learned that the bottleneck is almost never the research question. It is the access.

Doing Research On Crime And Justice requires the right databases and a realistic timeline

Start with the Bureau of Justice Statistics website if you need national-level figures. The data goes back decades but the release schedule is unpredictable. You will wait four to nine months for finalized annual reports after the reference year closes. If you need current numbers, check the Uniform Crime Reporting program though their methodology changed in 2021 and older comparisons break down. For academic work, jump straight into PsycINFO and Criminology and Criminal Justice Abstracts. Web of Science covers the high impact journals but misses a lot of state level research published in regional law reviews. Scopus is better for international coverage but the English language bias is real.

The National Archive of Criminal Justice Data gives you original survey and arrest data from government studies. The download process is slower than it should be. You request access, fill out CITI training modules, and wait two weeks minimum before they push the file through. Factor that in before you commit to a thesis topic.

What beginners consistently get wrong

They treat offense counts as absolute truth. They are not. Arrest data reflects police activity patterns, not actual crime rates. A city that increases patrol hours in one neighborhood will show more arrests there regardless of whether crime rose or fell. I saw this firsthand when a county sheriff switched to proactive policing in a low income area and the arrest numbers tripled overnight while the victimization survey showed zero change.

Another mistake is ignoring jurisdiction boundaries. County data, municipal data, state data, and federal data do not overlap cleanly. A report from 2019 might cover a city that dissolved its police department and merged with the county. The coding changes mid stream and your longitudinal analysis breaks unless you map those transitions manually.

Raw data access is harder than it sounds

The FBI does not publish microdata. Their Summary Reporting System gives aggregated counts and the National Incident Based Reporting System has a richer incident level dataset but researchers cannot download it directly. You have to apply through the NACJD or partner with an institution that already holds a license. The application review takes six to eight weeks on average.

State level repositories vary wildly. California publishes a lot. Mississippi publishes very little. I once spent three months trying to get arrest records from a southern state only to learn their system was not digitized past 2015. I pivoted to court records from the county clerk and rebuilt the timeline from there. That added four months but gave me individual case outcomes the state data never had.

Practical workflow that actually saves time

Build a codebook before you touch any data. I know that sounds obvious but most people skip it and then waste days guessing what a variable code meant. When I started working with police blotter data, the field labeled disposition had five different interpretations across three years. I coded each version into a master legend and flagged which fiscal year used which system. That saved me from running regression models on mismatched categories.

Use Python or R for cleaning. Excel breaks at around 500,000 rows and loses precision on large numeric IDs. I use pandas with strict data type enforcement and set up a validation step that flags records with impossible values. Missing race data shows up at a 12 percent rate in municipal arrest records. That is not random. It skews toward certain precincts. You need to note that and decide whether to impute or drop before it quietly biases your results.

Software options and where to find them

SPSS remains standard in criminal justice programs but it is expensive for individual researchers. JASP is free and handles most basic inferential tests without the licensing hassle. For spatial analysis, QGIS works fine for basic mapping though it lacks some of the advanced network analysis tools built into ArcGIS Pro. The university license usually covers that if you are a student.

Data visualization tools like Tableau Public let you publish interactive dashboards without paying. The catch is you cannot upload sensitive data. If your dataset contains individually identifiable information, keep it on a secure server and export anonymized aggregates instead. I learned that the hard way when a department reviewed my public dashboard and found resident addresses that had survived my deidentification step. It took a weekend to fix and another month to resubmit for IRB approval.

When your project hits a wall

You will hit this eventually. Your access request gets denied, a dataset turns out to be incomplete, or a key variable is coded differently than you expected. The workaround is usually building a secondary source as a backup from the start. If you are studying recidivism but the state prison dataset is incomplete, pull county jail release records as a proxy. They will not give you state level outcomes but they cover the six month window most projects care about.

I also recommend setting up a shared folder with your raw data, cleaned data, and analysis scripts in separate subfolders. Label everything with a date stamp and a version number. I lost two weeks once because I could not tell which cleaned file was the final version. The data had not changed but the variable labels had shifted and I was running the wrong model on the wrong file.

A note on ethics and IRB

If you are working with human subjects data even indirectly you need IRB approval. This includes many archival datasets where the information can be linked back to individuals. The NACJD data goes through a restricted use agreement that requires you to complete additional training. Without it, your project can be flagged later and publishers will ask for confirmation. Do not skip this step. It adds time but it prevents problems that are much harder to fix after submission.

Keep your consent documentation if you collect any primary data. A survey or interview project without proper consent forms will stall at review. I once had a participant withdraw after providing their statement and the department asked me to delete their record. I had the withdrawal form and the IRB approved amendment ready within two days. The paperwork itself is what made the difference.

Common sources you should check early

The National Vanishing Incarceration Project tracks jail closures and policy changes across states. It is not a dataset you analyze but it explains why certain years show sudden drops in custody populations. The MacArthur Foundation Research Network on Adolescent Development has longitudinal data that covers justice involvement as one outcome among many. Access is slower but the depth is useful for developmental criminology questions.

The Bureau of Labor Statistics provides occupational data if your research touches employment and criminal records. The link between a prior conviction and wage outcomes is well studied but the current data still gets underused. Department of Education civil rights data covers school discipline which feeds into the school to prison pipeline literature. Pulling those numbers early saves you from missing a key control variable later.

Final thoughts on the process

Research in this field is mostly about patience and documentation. The interesting findings come after you have gone through three rounds of cleaning and realized the first pass was wrong. I do not recommend rushing the data preparation phase. A properly cleaned dataset takes longer to build than most people expect but it pays off when you are writing the methods section and need to justify every variable transformation.

The field moves slowly because the data moves slowly. Budget your timeline accordingly and build in buffer time for access requests and approvals. The actual analysis is usually the fastest part of the project.