Transform DataFrame and compute diff-in-diff

You are given a pandas DataFrame df with the following columns:

• unit_id (string): entity identifier (e.g., user, city, driver)
• group (string): either 'treatment' or 'control'
• period (string): either 'pre' or 'post'
• y (string): outcome stored as a string (should be numeric), with exactly one missing value (NaN)

Tasks:

1. Convert y from string to integer (assume all non-missing values are valid integer strings, e.g. '12').
2. Impute the missing value in y using the simple (unconditional) average of the non-missing y values.
3. After steps (1)–(2), compute the difference-in-differences (DiD) estimate of the treatment effect on y:

$\text{DiD} = (\overline{y}_{\text{treat, post}} - \overline{y}_{\text{treat, pre}}) - (\overline{y}_{\text{ctrl, post}} - \overline{y}_{\text{ctrl, pre}})$

Return the scalar DiD estimate (and optionally the intermediate group-period means used).

Transform DataFrame and compute diff-in-diff

Table df

column	type	constraints
unit_id	VARCHAR	NOT NULL
group	VARCHAR	NOT NULL
period	VARCHAR	NOT NULL
y	VARCHAR

Sample data:

group	period	unit_id	y
treatment	pre	u1	10
treatment	post	u1	15
treatment	pre	u2	8
treatment	post	u2	14
control	pre	u3	9
control	post	u3	11
control	pre	u4	NULL
control	post	u4	10

Expected output

After casting y to integer and imputing the single missing y with the overall mean of non-missing values, return the DiD estimate along with the group-period means used.

ctrl_post_mean	ctrl_pre_mean	did_estimate	treat_post_mean	treat_pre_mean
10.5	10	5	14.5	9

Hints

1. CAST y from VARCHAR to INTEGER first; use AVG() which ignores NULLs to get the unconditional mean.
2. Impute with COALESCE(cast(y_int as decimal), global_avg). Bring the global average to every row via CROSS JOIN.
3. Quote the reserved column name "group" in SQL.
4. Compute group-period means, then pivot with conditional aggregation to form the DiD components.