Your data looks like it warrants a hash of hash rather than one-level hash structure; in other words it suggests something like:

my %HoH = (); # 'my' declares at file scope in this case
LoadFile(\%HoH); # passing by reference is more portable
                 # if you later make a module
Stats(\%HoH);

sub LoadFile{
    my $href = shift; # get the actual parameter
    open my $fileHandle, "<filepath" or die $!;
    while( <$fileHandle> {
        my ($st,$to,$te) = split( /\:/ );
        $href -> { $st }{ $to } = $te; # load file data into the HoH
    }
    close $fileHandle;
}

sub Stats {
    my $href = shift;
    for my $st ( sort keys %$href ) { # in state order
        my $toref = $href -> { $st }; # ref. to town subhash
        my ( $count, $sum ) = (0,0);
        for my $to ( keys %$toref ) { # sum through towns
                                      # (any order will do)
            $count++; $sum += ( $toref -> { $to } );
                # accumulate state aggregates
        }
        print "$st: " . $sum/$count . "\n";
            # average for state = sum/count
    }
}
[download]