Intergrated Labour Force Survey 2000-2001

Manual editing and coding

The Editing Manual was prepared and a special training session of five days was conducted from 24th July to 28th July 2000 to train editors.

A team of eight persons including subject matter officers was formed to undertake manual editing and coding of the questionnaires received from the regions. A special procedure for receiving questionnaires from the regions was set up. One person was assigned the task of recording the questionnaires as they arrived from different regions.

The ILFS questionnaire was very complex and required considerable diligence in editing and coding.

Major tasks of editors and coders were: -

To check if all questionnaires of a household were present.
To check each questionnaire for completeness, and make imputation if necessary.
To code for TASCO, ISIC and Subject of Training in all relevant questions.
To check “other” categories in questions to see if they could be re-allocated to a specific category.
To check sequence, completeness and consistency in both Labour Force (1990/91 and 2000/01) general questionnaire and child labour questions.

In general the manual editors and coders did not do their work well. Many errors were spotted later during computer edits. As a result the Computer editors took considerable time cleaning the data.

Frequencies and Checks of Data

Frequency tables of all the questions were prepared after the completion of data cleaning activity for each Quarter. These frequency tables assisted in locating and detecting extra errors, which were relatively much fewer. Although using QUICKTAB - a component of
IMPS application could produce frequency tables, however, all tables were produced using an SPSS package.

Extensive series of computer edits using IMPS-CONCOR a component of IMPS application was found to be more efficient and time saving than other types of edits. Several cycles were necessary before each quarter was clear of errors. Editing on the screen was faster and enabled one to edit several types of errors at once. Frequencies of all questions and table checks were run as soon as data cleaning was over. This facilitated locating unusual values, which were not picked up by computer edits.

Table 8.2: Summary of some results obtained from the frequency tables of the cleaned data for the four quarters
Descriptions Q1 Q2 Q3 Q4
No. of clusters involved 133 132 130 127
No. of House-holds interviewed 2,982 2,984 2,892 2,824
Fully responding House-holds 2,963 2,874 2,757 2,664
Not fully responded 119 110 135 160
No. of Household members 16,256 16,261 15,478 14,744
Average Household size 5.68 5.66 5.61 5.53
Percentage non- response 4.00 3.70 4.70 5.70

Some unusual results that were spotted are as follows:

There were “professionals “ with average monthly income as low as shs. 3,500/=
Persons employed in Central /Local Government with Monthly income below shs. 1000/= were spotted.
All the unusual values were crosschecked on the original questionnaires and necessary changes were made to rectify them.

8.5 Data Processing and Tabulation

Tabulation was done using a statistical package called SPSS. This package stores data files in fixed record length formats. Information from different sections of the survey questionnaires was stored in one ASCII file. Record types distinguish the different sections that had varying record lengths. So the ASCII file was divided into eight ASCII files each with a fixed record length. These were later exported into SPSS format together with the sampling weight file.

Note that under the recommendations of the computer experts, the SPSS and not the CENTS - a component of IMPS software was applied in the tabulation and analysis due to the problem of round off of the weighting program. Using SPSS minimized this problem. Some of the Tables produced in SPSS needed some improvements. Therefore they were copied into MS EXCEL worksheets and customized.

The process of deriving variables and merging files required for tabulation was done using an SPSS program (called lfs01-9.sps). Problems of missing values were encountered in derived variables. Modifying and improving the definitions was solved by the subject matter experts. In addition to that some variables had many categories or values which generated long tables. This problem was solved by deriving variables that group together the individual categories or variables.