Mandat mater 5,6,7 cleaning, recode dan compute

Main Ideas and Lessons

• Course context & logistics

  • The lecturer, Nurul Setyaningtyas, introduces the next topics in the data management course as preparation for the midterm exam.
  • Materials and assignments will be shared on Google Classroom, where students can ask questions.
  • The lecturer emphasizes that students must still complete assignments, and submission is tied to SPSS work.

• Data cleaning (cleansing)

  • Purpose: checking and ensuring the correctness of data entry.
  • What it checks for:
    • Validity of codes/values
      • Example: education code should only fall within an allowed range. Even if a total looks correct (e.g., total = 100), an invalid code (e.g., education code "7" when only 1–4 are allowed) indicates an entry error that must be re-checked.
    • Consistency across related variables/tables
      • Example: if only 20 respondents are KB participants, but a contraceptive-type frequency table sums to 24, then there are likely data entry errors because totals don’t logically match.
    • Plausibility checks using cross-tabulation
      • Example concept: comparing age vs. number of children can reveal impossible/unlikely combinations (e.g., a 15-year-old with a very high number of children). This suggests values may have been entered into the wrong case/value and require correction.

• Data quality criteria for successful cleansing

  • The lecturer lists key quality requirements, including:

  • Validity (accuracy of meaning and allowed structure)

    • Data must have clear types, allowed value ranges, no empty fields (or a decision strategy for empties), uniqueness, and proper formatting/patterns.

  • Completeness

    • Data elements should be filled. Empty data must be handled by either:
      • discarding the record/sample (if acceptable), or
      • filling missing values using additional information.
    • Completeness matters even for fields not used in analysis (e.g., names), because incomplete/duplicate/mistyped fields reflect input errors.

  • Consistency

    • Related variables must agree across the dataset (e.g., KB participant counts must match related contraceptive distributions).

  • Uniformity

    • Data should follow consistent standards (e.g., using the same unit/type representation throughout, such as “one size” consistently).

  • Additional criteria mentioned

    • Effectiveness & efficiency (time and project cost)
    • Data security
    • Data processing contingencies (handling risks during processing)

• Data transformation

  • Purpose: convert existing variables into new variables that better match analysis needs.

  • Typical transformation goals:

    • Grouping/recoding variables into categories
      • Example: convert numeric age into categories (e.g., young vs. risk vs. not at risk).
    • Combining variables with mathematical operations
      • Example: compute differences or totals (e.g., weight before vs. after; calculate change; compute HB increase/decrease).

  • SPSS terminology: “Compute” vs. “Recode/Record”

    • Compute (computer variable)

      • Creates or modifies values of a new variable using arithmetic or function-based operations.
      • Can work for numeric/string types, with emphasis on numeric.
      • SPSS workflow:
        1. Go to Transform → Compute Variable
        2. Choose the Target Variable (new variable name)
        3. Build the numeric expression (e.g., remainder = savings - loans, or HB_difference = HB2 - HB1)
        4. Confirm with OK
      • Example logic described:
        • difference calculation (e.g., HB2 - HB1) producing positive/negative/zero outcomes.

    • Recode into different variables (lecturer’s term for SPSS action analogous to recoding/grouping)

      • Used to group values into categories and store results in a new variable (recommended instead of overwriting the original).
      • SPSS workflow:
        1. Go to Transform → Recode into Different Variables
        2. Select the input variable
        3. Enter the output variable name (new variable)
        4. Define categories: old value ranges → new coded values
        5. Press Continue / OK
      • Example classification:
        • Baby weight:
          • LBW if < 2500 → code 1
          • Not LBW if ≥ 2500 → code 2
      • Lecturer emphasizes why automated recoding is better than manual:
        • manual grouping increases error rate, especially with large datasets and fatigue/time pressure.

• End of session / assignment reminder

  • The lecturer re-emphasizes that compute and recode (and/or grouping using SPSS transforms) will be used later in research.
  • Students are reminded to finish remaining assignments, with submissions handled via the classroom/assignment section.

---

Methodology / Instruction-Style Steps

A) Data Cleaning Approach (Checking Data Correctness)

1. Verify allowed value ranges/codes

   • Check categorical codes fall within the expected set (e.g., education code should be only 1–4).

2. Check totals and cross-table consistency

   • Compare related tables (e.g., KB participant counts vs. contraceptive-type totals).
   • If totals don’t match logically, identify likely entry errors.

3. Validate cross-tab plausibility

   • Use pairings such as age vs. number of children to find impossible/unlikely patterns.
   • Correct entries likely mis-entered into the wrong case/value.

---

B) SPSS “Compute Variable” Workflow (Math-based new variable)

1. In SPSS, go to Transform → Compute Variable
2. Set the Target Variable (name for the computed variable)
3. Choose/create the Numeric Expression
   • Examples mentioned:
     • remainder = savings - loans
     • HB_difference = HB2 - HB1
4. Click OK to apply the computation
5. Use the resulting variable for later analysis (differences may be negative/zero/positive)

---

C) SPSS “Recode into Different Variables” Workflow (Category grouping)

1. Go to Transform → Recode into Different Variables
2. Select the Input Variable
3. Set the Output Variable
   • Prefer creating a different/new variable so the original remains visible.
4. Define recoding rules
   • Specify old value range → new code value
   • Example:
     • if baby weight < 2500 → code 1 (LBW)
     • if baby weight ≥ 2500 → code 2 (Not LBW)
5. Click Change to add each rule
6. Click Continue, then OK
7. Use the new coded variable in analysis (e.g., value labels/categories)

---

Speakers / Sources Featured

• Nurul Setyaningtyas — lecturer/instructor for the data management course