ustat_var.varcovar

ustat_var.varcovar(origX, origY, w=None, quiet=True)

U-stat estimator of variance / covariance for teacher effects X and Y are a J-by-\(\operatorname{max}(T_j)\) matrix of teacher-specific mean residuals. When X and Y are residuals for the same outcome and covariate group, code will return a estimate of variance of teacher effects. When X and Y differ (either in outcome or Xs), code returns an estimate of the covariance.

Each row of X and Y are residuals for a specific teacher, ordered as first year observed, second year observed, etc. Since teachers have different number of years observed, X and Y should be np.NaN for all years after the last year observed. Each teacher must have at least 2 years observed.

X and Y must have the same dimension.

Note: this is a wrapper function that calls varcovar_balanced or varcovar_ustat depending on whether the panels are balanced or unbalanced.

“varcovar_balanced.py” is used when the panels are balanced, i.e. each teacher appears the same number of times in both X and Y.

“varcovar_ustat.py” is used when the panels are unbalanced, i.e. each teacher appears a different number of times in X and Y.

origX: array

J-by-\(\operatorname{max}(T_j)\) array containing residuals/data for outcome X

origY: array

J-by-\(\operatorname{max}(T_j)\) array containing residuals/data for outcome Y

w: array

(Optional) J-by-1 array of user-supplied weights. If supplied, varcovar will return row-weighted variance-covariance of row means.

quiet: boolean

(Optional) If quiet=True, function call will report type of variance being calculated (unweighted/weighted) and whether panels are balanced or unbalanced.

float

Variance-covariance between rowmeans of origX and origY.