7

I am having difficulty understanding how to correctly transpose this table from a database I am using. This table was created by someone who designed the database years ago, but I managed to retrieve the table's create statement using pg_dump.

Here's the table with an example entry:

CREATE TABLE response (
    session_id integer NOT NULL,
    seconds integer NOT NULL,
    question_id integer NOT NULL,
    response character varying(500),
    file bytea
);

INSERT INTO response(session_id, seconds, question_id, response, file)
VALUES (758,1459505869,31,'0',''),  (758,1459505869,32,'0',''), 
(758,1459505869,33,'0',''), (758,1459505869,34,'0',''), 
(758,1459505869,35,'1',''), (758,1459505869,36,'0',''), 
(758,1459505869,37,'0',''), (758,1459505869,38,'0',''), 
(758,1459506973,38,'0',''), (758,1459506973,37,'0',''), 
(758,1459506973,36,'0',''),(758,1459506973,35,'1',''),  
(758,1459506973,34,'0',''),(758,1459506973,33,'0',''),  
(758,1459506973,32,'0',''),(758,1459506973,31,'0',''),
(758,1459508676,31,'0',''),(758,1459508676,32,'0',''),  
(758,1459508676,33,'0',''),(758,1459508676,34,'0',''),  
(758,1459508676,35,'1',''),(758,1459508676,36,'0',''),  
(758,1459508676,37,'0', ''),    (758,1459508676,38,'0', '');

SELECT * FROM response LIMIT 5;
session_id  seconds   question_id   response    file
   758     1459505869     31           0         [null]
   758     1459505869     32           0         [null]
   758     1459505869     33           0         [null]
   758     1459505869     34           0         [null]
   758     1459505869     35           1         [null]

The question id in the question_id column stands for the following:

30  -- not_foot_count 
31  -- not_moving
32  -- foot
33  -- bicycle
34  -- motorcycle
35  -- car
36  -- bus
37  -- metro
38  -- other
39  -- train

Response can be text (wrong user response), but mostly a 1 or a 0 (which are of interest to me).

So I want to transpose this table into a new table survey, so that the returned query result will have for each column, the corresponding response code value as the column name (32 -> foot; 33 -> bike; 35 -> car. etc)

I am not interested in all these responses, but 5: foot, bike (for bicycle), bus, car, and metro.

Because I had great trouble retrieving only the 5 responses of interest, I started with retrieving all this values to see if I am doing things correctly. It turns out that I am doing things wrong.

Here's my attempt:

CREATE TABLE survey
AS
SELECT aresult.session_id,
   aresult.not_foot_count,
    aresult.not_moving,
    aresult.foot,
    aresult.bike,
    aresult.motor,
    aresult.car,
    aresult.bus,
    aresult.metro,
    aresult.train,
    aresult.other
   FROM crosstab('select session_id, question_id, response 
                from  response
                order by session_id,question_id'::text) 
  aresult(session_id integer, not_foot_count character varying(500), 
  not_moving character varying(500), foot character varying(500), 
  bike character varying(500), motor character varying(500), 
  car character varying(500), bus character varying(500), 
  metro character varying(500), train character varying(500), 
  other character varying(500));

Which gives:

SELECT * FROM survey;
session_id seconds not_foot_count not_moving foot bike motor car bus metro train other
758       1459505869       0          0        0   0    0     0   0   0      0     0
758       1459506973       0          0        0   0    0     0   0   0      0     0
758       1459508676       0          0        0   0    0     0   0   0      0     0

Notice that this is not correct as the column car should be 1.

Besides, I am not not interested in all the values. Instead would like only values of interest to be.

Expected out

I expect to limit my return result to the following (with the correct answer):

session_id  seconds    foot bike car bus metro 
    758   1459505869    0     0   1   0    0
    758   1459506973    0     0   1   0    0
    758   1459508676    0     0   1   0    0

Note: My attempt is illustrated in this dbfiddle.

EDIT

As far comments, question edited to show complete expected output.

6
  • What is supposed to be done with all of those different seconds?
    – jjanes
    Sep 27, 2022 at 2:02
  • Unrelated to your problem, but: Postgres 9.6 is no longer supported you should plan an upgrade as soon as possible.
    – user1822
    Sep 27, 2022 at 5:54
  • @jjanes Although not used here, the seconds represent the response submission time (unix).
    – arilwan
    Sep 27, 2022 at 7:13
  • @a_horse_with_no_name OK, removing the tag. But the question is generally about the building sql to return the needed result set.
    – arilwan
    Sep 27, 2022 at 7:14
  • You have two responses for question_id = 36. If one would be one and the other two (instead of both being 0 in your sample data) what should the output for bus be then? A comma separated list of the responses, e.g. one,two?
    – user1822
    Sep 27, 2022 at 7:21

1 Answer 1

4

I am not a fan of the crosstab() function as I find it more complicated than filtered aggregation (and it doesn't workaround the fact that you need to specify all result columns manually).

The following returns what you want.

select session_id, 
       seconds, 
       max(response) filter (where question_id = 32) as foot,
       max(response) filter (where question_id = 33) as bike,
       max(response) filter (where question_id = 36) as bus,
       max(response) filter (where question_id = 35) as car,
       max(response) filter (where question_id = 37) as metro
from response
group by session_id, seconds
2
  • Thank you, this is even easily readable. I appreciate.
    – arilwan
    Sep 27, 2022 at 9:08
  • Elegant, precise solution. You make my day.
    – arilwan
    Sep 27, 2022 at 9:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.