--- configs: - config_name: default data_files: - split: train path: data/train-* dataset_info: features: - name: lsat dtype: float32 - name: gpa dtype: float32 - name: race dtype: class_label: names: '0': Asian '1': Black '2': Hispanic '3': White - name: resident dtype: class_label: names: '0': out-of-state '1': in-state - name: college dtype: class_label: names: '0': Akron '1': Arizona '2': Arizona State '3': Baltimore '4': Buffalo '5': Cincinnati '6': Cleveland State '7': George Mason '8': Hawaii '9': Houston '10': Idaho '11': Louisiana State '12': Michigan '13': Minnesota '14': Missouri at Columbia '15': Missouri at Kansas City '16': Nevada Las Vegas '17': North Carolina '18': Northern Illinois '19': Ohio State '20': Virginia '21': Washington '22': West Virginia '23': William and Mary '24': Wyoming - name: year dtype: class_label: names: '0': '2005' '1': '2006' '2': '2007' '3': '2008' '4': 200x - name: gender dtype: class_label: names: '0': female '1': male - name: admit dtype: class_label: names: '0': denied '1': admitted - name: black dtype: class_label: names: '0': 0.0 '1': 1.0 - name: hispanic dtype: class_label: names: '0': 0.0 '1': 1.0 - name: asian dtype: class_label: names: '0': 0 '1': 1 - name: white dtype: class_label: names: '0': 0 '1': 1 - name: missingrace dtype: class_label: names: '0': 0 '1': 1 - name: urm dtype: class_label: names: '0': 0 '1': Black OR Hispanic - name: enroll dtype: class_label: names: '0': not enrolled '1': enrolled splits: - name: train num_bytes: 14090514 num_examples: 124557 download_size: 1062910 dataset_size: 14090514 --- # Law School Admission Dataset ## Dataset Overview The **Law School Admission Dataset** provides detailed application and admission records from **25 law schools** for the **2005, 2006, and, in some cases, 2007 and 2008 admission cycles**. The dataset includes over **100,000 individual applications** and contains variables related to **academic performance, demographics, residency, and admission decisions**. ## Dataset Structure - **Number of examples:** 124,557 - **Number of features:** 14 - **Dataset size:** 13.98 MB (compressed) - **Download size:** 1.08 MB ## Features | Feature Name | Type | Description | |--------------|-----------|--------------| | `lsat` | `float32` | LSAT score (scaled from **120 to 180**). Some schools used a different scale, which has been recoded. | | `gpa` | `float32` | Undergraduate GPA on a **4-point scale**. | | `race` | `categorical` | Self-reported race: **Asian, Black, Hispanic, White, or missing**. | | `resident` | `categorical` | Residency status: **in-state (1)** or **out-of-state (0)**. | | `college` | `categorical` | Undergraduate institution (aggregated to **25 unique institutions**). | | `year` | `categorical` | Intended enrollment year: **2005, 2006, 2007, 2008, or suppressed as 200x**. | | `gender` | `categorical` | **Male (1)** or **Female (0)**. | | `admit` | `categorical` | Admission decision: **admitted (1)** or **denied (0)**. | | `black` | `binary` | Indicator variable for **Black applicants** (1 = Black, 0 = not Black). | | `hispanic` | `binary` | Indicator variable for **Hispanic applicants** (1 = Hispanic, 0 = not Hispanic). | | `asian` | `binary` | Indicator variable for **Asian applicants** (1 = Asian, 0 = not Asian). | | `white` | `binary` | Indicator variable for **White applicants** (1 = White, 0 = not White). | | `missingrace` | `binary` | Indicator for missing race information. | | `urm` | `binary` | **Underrepresented minority (URM):** **1 if Black OR Hispanic, 0 otherwise**. | | `enroll` | `categorical` | Enrollment decision **conditional on admission**: **1 = enrolled, 0 = not enrolled or deferred**. | ## Data Collection & Processing - **Source:** Law school application records for 25 institutions. - **Privacy Measures:** To minimize re-identification risks: - Small cell sizes (<5 individuals) were aggregated. - **Black and Hispanic applicants were grouped as URM** in some cases. - Gender and year variables were **suppressed where necessary** to maintain anonymity. - **Data Cleaning:** - **LSAT scores recoded** for schools that used non-standard scales. - **Outliers removed** (erroneous data entries). ## Dataset Splits - **Train split:** 124,557 records - **Configs:** Single configuration (`default`) - **Data Files:** Stored in the `data/train-*` format. ## Intended Uses & Applications This dataset is useful for: - **Law school admission modeling** - **Fairness and bias analysis** in admissions decisions - **Predictive analytics** for law school acceptance - **Educational policy research** ## Limitations & Considerations - **Aggregated and suppressed categories** may limit granular subgroup analysis. - **Only includes applicants from 25 schools**, not representative of all U.S. law schools. - **Admission decisions** do not capture subjective evaluation factors (e.g., essays, recommendations). ## Citation If you use this dataset, please consider cite the following works (these might be inaccurate, but note that the publication by Linda F. Wightman (especially the 1990s version) appears to be something else): ```bibtex @article{smyth2004ethnic, title={Ethnic and Gender Differences in Science Graduation at Selective Colleges with Implications for Admission Policy and College Choice}, author={Smyth, Frederick L. and McArdle, John J.}, journal={Research in Higher Education}, volume={45}, number={4}, pages={353--381}, year={2004}, publisher={Springer} } @article{arcidiacono2007representation, title={Representation versus Assimilation: How do Preferences in College Admissions Affect Social Interactions?}, author={Arcidiacono, Peter and Khan, Shakeeb and Vigdor, Jacob L.}, journal={Working Paper}, year={2007}, institution={Duke University} } ``` ## The source Unfortunately, the source of this data is still a myth. This could by no means be the original data by [Wightman](https://files.eric.ed.gov/fulltext/ED469370.pdf) or [Richard Sander](https://www.brown.edu/Departments/Economics/Faculty/Glenn_Loury/louryhomepage/teaching/Ec%20137/Richard%20Sander%20on%20Affirmative%20Action%20in%20Law%20Schools.pdf), as these two papers were published before 2004. We roughly remember we came across a web page with the two papers in bibtex above, which could be the right page. Sadly, we cannot remember where it is, nor can we find it again. We really hope you can let us know the real source if you happen to know it. We used the following code to produce this Hugging Face dataset. ``` wget https://web.archive.org/web/20210427055113/http://www.seaphe.org/databases/FOIA/lawschs1_0.dta wget https://web.archive.org/web/20210427055113/http://www.seaphe.org/databases/FOIA/lawschs1_1.dta ``` (Note that we need to use this link. The original link is gone. Also, ```lawschs1_0.dta``` and ```lawschs1_1.dta``` only have < 5000 different rows out of > 120 k rows (if you happen to see number 1507 or 4556, then you are even closer). So we would just use ```lawschs1_1.dta```.) ```python from datasets import Dataset, DatasetDict, Features, Value, ClassLabel # Define categorical and continuous columns continuous_columns = ["lsat", "gpa"] categorical_columns = ["race", "resident", "college", "year", "gender", "admit", "black", "hispanic", "asian", "white", "missingrace", "urm", "enroll"] # Convert categorical columns to category type and get mappings category_mappings = {} for col in categorical_columns: # df_law[col] = df_law[col].astype("category") category_mappings[col] = df_law[col].astype("category").cat.categories.to_list() # df_law[col] = df_law[col].cat.codes # Convert to integer codes # Define Hugging Face Dataset Features hf_features = Features({ "lsat": Value("int64"), "gpa": Value("float32"), "race": ClassLabel(names=category_mappings["race"]), "resident": ClassLabel(names=['out-of-state', 'in-state']), "college": ClassLabel(names=category_mappings["college"]), "year": ClassLabel(names=[str(y) for y in category_mappings["year"]]), "gender": ClassLabel(names=["female", "male"]), "admit": ClassLabel(names=["denied", "admitted"]), "black": ClassLabel(names=category_mappings["black"]), "hispanic": ClassLabel(names=category_mappings["hispanic"]), "asian": ClassLabel(names=category_mappings["asian"]), "white": ClassLabel(names=category_mappings["white"]), "missingrace": ClassLabel(names=category_mappings["missingrace"]), "urm": ClassLabel(names=[0, 'Black OR Hispanic']), "enroll": ClassLabel(names=["not enrolled", "enrolled"]), }) # Convert to Hugging Face dataset hf_dataset = Dataset.from_pandas(df_law, features=hf_features) # Create DatasetDict for organization hf_dataset_dict = DatasetDict({"train": hf_dataset}) # Display dataset structure hf_dataset_dict ```