auc_compare {auctestr}

Compare AUC values using the FBH method.

Description

Apply the FBH method to compare outcome_col by compare_col, averaging over time_col (due to non-independence) and then over over_col by using Stouffer's Method.

Usage

auc_compare(df, compare_values, filter_value, time_col = "time",
  outcome_col = "auc", compare_col = "model_id", over_col = "dataset",
  n_col = "n", n_p_col = "n_p", n_n_col = "n_n",
  filter_col = "model_variant")

Arguments

df

DataFrame containing time_col, outcome_col, compare_col, and over_col.

compare_values

names of models to compare (character vector of length 2). These should match exactly the names as they appear in compare_col.

filter_value

(optional) keep only observations which contain filter_value for filter_col.

time_col

name of column in df representing time of observations (z-scores are averaged over time_col within each model/dataset due to non-independence). These can also be other dependent groupings, such as cross-validation folds.

outcome_col

name of column in df representing outcome to compare; this should be Area Under the Receiver Operating Characteristic or A' statistic (this method applies specifically to AUC and not other metrics (i.e., sensitivity, precision, F1)..

compare_col

name of column in df representing two conditions to compare (should contain at least 2 unique values; these two values are specified as compare_values).

over_col

identifier for independent experiments, iterations, etc. over which z-scores for models are to be compared (using Stouffer's Z).

n_col

name of column in df with total number of observations in the sample tested by each row.

n_p_col

name of column in df with n_p, number of positive observations.

n_n_col

name of column in df with n_n, number of negative observations.

filter_col

(optional) name of column in df to filter observations on; keep only observations which contain filter_value for filter_col.

Value

numeric, overall z-score of comparison using the FBH method.

References

Fogarty, Baker and Hudson, Case Studies in the use of ROC Curve Analysis for Sensor-Based Estimates in Human Computer Interaction, Proceedings of Graphics Interface (2005) pp. 129-136.

Stouffer, S.A.; Suchman, E.A.; DeVinney, L.C.; Star, S.A.; Williams, R.M. Jr. The American Soldier, Vol.1: Adjustment during Army Life (1949).

See Also

Other fbh method: fbh_test, se_auc

Examples

## load sample experiment data
data(sample_experiment_data)
## compare VariantA of ModelA and ModelB
auc_compare(sample_experiment_data,
    compare_values = c('ModelA', 'ModelB'),
    filter_value = c('VariantA'),
    time_col = 'time',
    outcome_col = 'auc',
    compare_col = 'model_id',
    over_col = 'dataset',
    filter_col = 'model_variant')
## compare VariantC of ModelA and ModelB
auc_compare(sample_experiment_data,
    compare_values = c('ModelA', 'ModelB'),
    filter_value = c('VariantC'),
    time_col = 'time',
    outcome_col = 'auc',
    compare_col = 'model_id',
    over_col = 'dataset',
    filter_col = 'model_variant')
## compare ModelC, VariantA and VariantB
auc_compare(sample_experiment_data,
    compare_values = c('VariantA', 'VariantB'),
    filter_value = c('ModelC'),
    time_col = 'time',
    outcome_col = 'auc',
    compare_col = 'model_variant',
    over_col = 'dataset',
    filter_col = 'model_id')