Aggregated results without duplicates on multiple columns each filtered differently

user_13085106 0
Viewed 124

I have a BigQuery table that looks like this and that can't be modified:

Country Customer Number of connections Number of purchases Country Metric 1 Country Metric 2
Brazil A 10 1 3 1000
Brazil B 90 5 3 1000
Brazil C 80 2 3 1000
Namibia B 20 1 5 2000
Namibia C 150 2 5 2000

About this table, please note that:

  • Each combination of Country-Customer is unique.
  • The country metrics, as their names suggest, only depend on the country.
  • For some countries, some metrics are not available (NULL in the table).
  • For some combinations Country-Customer, the number of connections/purchases are not available

I would like to obtain, in the same query, the following information:

  • The mean of Country Metric 1 only taking into account combinations of Country-Customer having at least a number of purchases higher or equal than 2. In the example table, there are 3 combinations: Brazil-B, Brazil-C and Namibia-C. The mean should take into account Brazil only once so the result is (3 + 5) / 2 = 4.
  • The mean of Country Metric 2 only taking into account combinations of Country-Customer having at least a number of connections higher than 100. There is only one combination which meets this criterion in the example table: Namibia-C. Thus, the expected result is 2000.

Those are just examples but there can be more metrics and other aggregations (sum, min, max, count...) but they should be very similar.

Here is what I've tried:

SELECT AVG(IF(purchases > 2, country_metric_1, NULL)), -- => 6.5
AVG(IF(connections > 100, country_metric_2, NULL)) -- => 2000
FROM table

Issue: if the same country appears in multiple combinations, the same metric is taken into account multiple times.

SELECT AVG(IF(purchases > 2, country_metric_1_p, NULL)), -- => random
AVG(IF(connections > 100, country_metric_2_p, NULL)) -- => random
FROM (SELECT purchases, 
connections,
IF(ROW_NUMBER() OVER (PARTITION BY country) = 1, country_metric_1, NULL) country_metric_1_p
IF(ROW_NUMBER() OVER (PARTITION BY country) = 1, country_metric_2, NULL) country_metric_2_p
FROM table)

Issue: for each country, only one combination is taken into account giving lower and random results...

Is there a way to achieve this?

google-bigquery sql
1 Answers

With this query:

SELECT MAX(CASE WHEN purchases >= 2 THEN country_metric_1 END) country_metric_1,
       MAX(CASE WHEN connections > 100 THEN country_metric_2 END) country_metric_2
FROM tablename 
WHERE purchases >= 2 OR connections > 100
GROUP BY country

you get the metrics that you want once for each country.
Use the above query as a CTE and average its results:

WITH cte AS (
    SELECT MAX(CASE WHEN purchases >= 2 THEN country_metric_1 END) country_metric_1,
           MAX(CASE WHEN connections > 100 THEN country_metric_2 END) country_metric_2
    FROM tablename 
    WHERE purchases >= 2 OR connections > 100
    GROUP BY country 
)
SELECT AVG(country_metric_1) avg_country_metric_1,
       AVG(country_metric_2) avg_country_metric_2
FROM cte

See the demo.
Results:

avg_country_metric_1 avg_country_metric_2
4 2000
user_104988280
Related