So long as your subset is a nested hashref or nested array, this works nicely. I've done something similar to spread processing across dedicated routines.

    my $results = process_planet( \%world );

    sub process_planet {
        my ( $world ) = @_;
        my @retval;
        for my $country_id ( keys %$world ) {
            my $population = process_country( $world->{$country_id} ) 
+;
            push @retval, { $country_id => $population };
        }
        return \@retval;
    }

    sub process_country {
        my ( $country ) = @_;
        my $national_pop;
        for my $city ( keys %$country ) {
            $national_pop += extract_city_population( $countr->{$city}
+ );
        }
        return $national_pop;
    }
[download]

Of course, at eavh level you have to know what you're receiving, and what you're returning.