We analyzed the spatial distribution of 28500 photometrically selected galaxies with magnitude 23.5 < { \cal R } _ { AB } < 25.5 
and redshift 1.4 < z < 3.5 in 21 fields with a total area of 0.81 square degrees .
The galaxies were divided into three subsamples , with mean redshifts \bar { z } = 1.7 , 2.2 , 2.9 , according to the U _ { n } G { \cal R } 
selection criteria of Adelberger et al .
( 2004 ) and Steidel et al .
( 2003 ) .
Combining the galaxies ’ measured angular clustering with redshift distributions inferred from 1600 spectroscopic redshifts , we find comoving correlation lengths at the three redshifts of r _ { 0 } = 4.5 \pm 0.6 , 4.2 \pm 0.5 , and 4.0 \pm 0.6 h ^ { -1 } Mpc , respectively , and infer a roughly constant correlation function slope of \gamma = 1.6 \pm 0.1 .
We derive similar numbers from the 1600 object spectroscopic sample itself with a new statistic , K , that is insensitive to many possible systematics .
Galaxies that are bright in { \cal R } 
( \lambda _ { rest } \sim 1500 – 2500 Å ) cluster more strongly than fainter galaxies at z = 2.9 and z = 2.2 but not , apparently , at z = 1.7 .
Comparison to a numerical simulation that is consistent with recent WMAP observations suggests that galaxies in our samples are associated with dark matter halos of mass 10 ^ { 11.2 } – 10 ^ { 11.8 } M _ { \odot } ( z = 2.9 ) , 10 ^ { 11.8 } – 10 ^ { 12.2 } M _ { \odot } ( z = 2.2 ) , 10 ^ { 11.9 } – 10 ^ { 12.3 } M _ { \odot } ( z = 1.7 ) , and that a small fraction of the halos contain more than one galaxy that satisfies our selection criteria .
Adding recent observations of galaxy clustering at z \sim 0 and z \sim 1 
to the simulation results , we conclude that the typical object in our samples will evolve into an elliptical galaxy by redshift z = 0 and will already have an early-type spectrum by redshift z = 1 .
We comment briefly on the implied relationship between galaxies in our survey and those selected with other techniques .