Dataset Viewer
Auto-converted to Parquet Duplicate
table_id
stringlengths
10
10
family
stringclasses
3 values
n
int64
500
500
n_covariates
int64
0
20
truth
float64
-1.84
1.74
archive_path
stringlengths
27
27
train_0000
qiu
500
10
1.119713
train/tables/train_0000.npz
train_0001
ding
500
0
0.364117
train/tables/train_0001.npz
train_0002
ding
500
0
0.034774
train/tables/train_0002.npz
train_0003
qiu
500
2
-0.781991
train/tables/train_0003.npz
train_0004
kennedy2020
500
1
0.110269
train/tables/train_0004.npz
train_0005
qiu
500
5
-0.886873
train/tables/train_0005.npz
train_0006
kennedy2020
500
1
-0.048509
train/tables/train_0006.npz
train_0007
kennedy2020
500
1
-0.104993
train/tables/train_0007.npz
train_0008
kennedy2020
500
1
-0.097024
train/tables/train_0008.npz
train_0009
kennedy2020
500
1
-0.124044
train/tables/train_0009.npz
train_0010
qiu
500
5
0.625687
train/tables/train_0010.npz
train_0011
kennedy2020
500
1
-0.342438
train/tables/train_0011.npz
train_0012
qiu
500
5
0.275321
train/tables/train_0012.npz
train_0013
ding
500
0
-0.399004
train/tables/train_0013.npz
train_0014
ding
500
0
-0.168122
train/tables/train_0014.npz
train_0015
qiu
500
5
-1.844466
train/tables/train_0015.npz
train_0016
ding
500
0
0.664718
train/tables/train_0016.npz
train_0017
qiu
500
20
0.817299
train/tables/train_0017.npz
train_0018
kennedy2020
500
1
0.178001
train/tables/train_0018.npz
train_0019
ding
500
0
0.037551
train/tables/train_0019.npz
train_0020
ding
500
0
-0.284091
train/tables/train_0020.npz
train_0021
ding
500
0
0.096503
train/tables/train_0021.npz
train_0022
ding
500
0
-0.082265
train/tables/train_0022.npz
train_0023
qiu
500
5
-0.051144
train/tables/train_0023.npz
train_0024
qiu
500
2
1.740879
train/tables/train_0024.npz
train_0025
kennedy2020
500
1
-0.124803
train/tables/train_0025.npz
train_0026
ding
500
0
-0.198298
train/tables/train_0026.npz
train_0027
ding
500
0
0.652
train/tables/train_0027.npz
train_0028
ding
500
0
0.166525
train/tables/train_0028.npz
train_0029
ding
500
0
-0.421613
train/tables/train_0029.npz
train_0030
qiu
500
20
1.36207
train/tables/train_0030.npz
train_0031
qiu
500
20
1.326563
train/tables/train_0031.npz
train_0032
kennedy2020
500
1
-0.116706
train/tables/train_0032.npz
train_0033
qiu
500
20
-0.41788
train/tables/train_0033.npz
train_0034
ding
500
0
-0.289305
train/tables/train_0034.npz
train_0035
kennedy2020
500
1
-0.280791
train/tables/train_0035.npz
train_0036
ding
500
0
0.379838
train/tables/train_0036.npz
train_0037
kennedy2020
500
1
0.010814
train/tables/train_0037.npz
train_0038
kennedy2020
500
1
0.305361
train/tables/train_0038.npz
train_0039
qiu
500
2
-1.240886
train/tables/train_0039.npz
train_0040
qiu
500
10
-0.281032
train/tables/train_0040.npz
train_0041
ding
500
0
-0.169695
train/tables/train_0041.npz
train_0042
kennedy2020
500
1
0.156938
train/tables/train_0042.npz
train_0043
kennedy2020
500
1
0.042098
train/tables/train_0043.npz
train_0044
kennedy2020
500
1
-0.266536
train/tables/train_0044.npz
train_0045
qiu
500
20
-1.740411
train/tables/train_0045.npz
train_0046
qiu
500
2
-0.453156
train/tables/train_0046.npz
train_0047
kennedy2020
500
1
-0.159369
train/tables/train_0047.npz

Statistical Inference Arena · Instrumental Variables (LATE)

Estimate the local average treatment effect (LATE) from a study with a binary instrument and a binary treatment, and return a two-sided 95% confidence interval. Each study contains 500 observations and contributes one estimate. The benchmark compares a frozen TabPFN-IV adapter checkpoint and a cross-fitted random forest on the same 2,000 studies. Model

Purpose Studies Download
Historical adapter fitting 20,000 Training archive
Historical checkpoint selection 1,000 Validation archive
Try estimation and scoring 48 train, 16 dev, 32 test Starter kit
Evaluate a fixed method 2,000 Observed inputs, folds and prediction template
Reproduce the comparison Same 2,000 Population LATE labels, scorer and reference predictions

Target and observations

For each study, hold its data-generating mechanism and parameters fixed. The target is the average effect of treatment among compliers, the units whose treatment follows the instrument.

LATE=E ⁣[Y(1)−Y(0)∣D(1)>D(0)]. \mathrm{LATE} = \mathbb{E}\!\left[Y(1)-Y(0)\mid D(1) > D(0)\right].

Here D(z) is a unit's treatment under instrument value z and Y(d) its outcome under treatment d. The expectation is over compliers in the population that generated the study. Every design satisfies exclusion, monotonicity (D(1) ≥ D(0)) and independence of the instrument given the covariates, so the LATE equals the ratio of the covariate-adjusted instrument effects on the outcome and on the treatment. Report one estimate in outcome units (a probability difference for binary outcomes) and one two-sided 95% confidence interval per study.

Each NPZ holds baseline covariates X, a binary instrument Z, a binary treatment D and the outcome Y, in matching row order. An array with shape (500, 0) denotes a study with no observed covariates. Supply only these four arrays to an estimator.

Run a first comparison

Extract the starter kit, install NumPy and run the following from the extracted directory.

python baseline.py dev/tables > predictions.csv
python score.py dev/truth.csv predictions.csv

The example computes the unadjusted Wald ratio (difference in mean outcomes over difference in mean treatment between the two instrument arms) with a delta-method normal interval. It ignores the covariates, so designs in which the instrument depends on X bias this simple estimate. On the 16 development studies it gives bias 0.4994, RMSE 1.1440, coverage 68.75% and mean interval length 0.7004. The starter draws 96 separate studies from the three-family prior with seed 20261119 and includes the generator (generate.py with iv_data.py and reference/; needs numpy and scipy). Its three viewer splits describe the files inside the starter ZIP.

Historical data and evaluation design

The historical files contain the original adapter training and validation studies. For each study the generator draws one of three families with equal probability (Ding 2017, Qiu 2021, Kennedy 2020), then the family's parameters from our prior, then 564 rows: 500 context rows followed by 64 query rows. Each NPZ also stores the analytic nuisance values at the query rows (targets: e(x), mD(x,0), mD(x,1), mY(x,0), mY(x,1)), the column permutation the adapter saw and metadata with the parameters, seeds and true LATE. Use only the observed arrays as model inputs. Root seed 20261116.

The evaluation uses one pooled split. For each study the generator draws one of the same three families uniformly, draws fresh parameters from the same prior, then samples 500 units. Seed 20261117 selects the exact tables used for the published comparison; the realized counts are Ding 665, Qiu 658 and Kennedy 677. There is no in-distribution / out-of-distribution split and one parameter draw per table, so the error variance is across studies. The input archive preserves the observed NPZ bytes and row order under opaque table IDs. The viewer's evaluation_inputs configuration lists one row per study with its size, number of covariates, outcome type and the SHA-256 of the input file. DGP families, priors and realized counts

Evaluate a method

Freeze the estimator before opening the scoring archive. The evaluation labels are public for reproducibility. Keep the supplied five folds for cross-fitting: one fixed random 5-fold split of the 500 rows, identical for every study and used by both reference methods. Row indices start at 0 and fold labels run from 1 to 5. Write one row per table using the template columns below.

table_id,estimate,ci_lower,ci_upper

Run python score.py truth.csv predictions.csv after extracting the scoring archive. Each study has equal weight. Bias and RMSE use estimate minus population LATE. Error variance uses the sample variance of those errors (denominator 1,999). Coverage counts intervals containing the truth, including either endpoint. Mean interval length averages upper minus lower. Measure elapsed seconds per study during estimation (point estimate and interval, including any bootstrap; excluding model loading and warm-up) and record the hardware if reporting runtime.

TabPFN-IV and the random-forest comparator estimate the instrument propensity e(x), the treatment means mD(x,z) and the outcome means mY(x,z) with the same 5-fold cross-fitting, then apply the same augmented inverse-probability-weighted ratio for the LATE. TabPFN-IV (bf16, one process on one NVIDIA L4) uses an influence-function interval. The random forest (scikit-learn, 200 trees, minimum leaf size 5, one single-threaded process per core on an AMD EPYC 7B13) uses 50 whole-row bootstrap resamples with refitting and a percentile interval; the trimmed variant clips e(x) to [0.05, 0.95] before the ratio. Their predictions, timings and the resulting metrics are in the scoring archive (predictions_pfn.csv, predictions_rf.csv, predictions_rf_trim.csv, reference_metrics.json):

Method RMSE Bias Error var. 95% coverage CI length Time (s/table)
TabPFN-IV 0.268 0.012 0.072 95.65% 0.875 0.366
RF + bootstrap 3.408 0.079 11.613 91.40% 5.390 182.6
RF (trimmed) + bootstrap 0.316 0.022 0.099 90.25% 1.111 179.9

The untrimmed forest fails on the Kennedy design, whose single covariate lets leaf frequencies reach a propensity of exactly 0 or 1; 260 of its 677 studies have an absolute error above 1.

References

  • Ding, VanderWeele and Robins (2017). Instrumental variables as bias amplifiers with general outcome and confounding. Biometrika 104(2), 291–302. Paper
  • Qiu, Tao and Zhou (2021). Inference of heterogeneous treatment effects using observational data with high-dimensional covariates. JRSS-B 83(5), 1016–1043. Paper
  • Kennedy, Balakrishnan and G'Sell (2020). Sharp instruments for classifying compliers and generalizing causal effects. Annals of Statistics 48(4). Paper
  • The Ding and Qiu generators are verbatim copies of the Arena registry files iv-ding-2017-binary-bias-v1.py and iv-qiu-2021-principal-stratum-poisson-v1.py in stats-inference-arena.
Downloads last month
32

Paper for Brownight/StatsInference-Arena-IV