Processing of data starts upon submission of the questionnaires to the office and ends with the compilation of these data into statistical tables. Processing of disability data follows the same process as that of any other data collected from survey, census or administrative-based data collection.
The processes involved are:
- Receipt and Control
- General Screening
- Manual checking of the questionnaires for consistency and completeness of entries
- Data encoding
- Computer check of geographic identification
- Consistency edit check
- Generation of preliminary tables
- Evaluation of preliminary tables
- Finalization of tables
Detailed instructions for those personnel who execute the above processes such as processors, editors, encoders, system analysts, programmers, and statisticians should be included in processing guides, separately for manual processing (from receipt and control to manual checking of questionnaires, and evaluation of tables) and computer processing (data encoding to generation of tables, and finalization of tables)
The brief description of each process is given below:
Receipt and control
This is the process of receiving the questionnaires from the enumerators/supervisors in all areas (or sample areas in the case of a survey). Upon receipt, the geographic identification of the questionnaires is recorded in a control book with signature of the person who received and transmitted the questionnaires to the office.
General screening is done by going over the submitted accomplished forms and checking for the completeness of the geographic identification such as region, province, municipality, village, and other information asked for in the cover page of the questionnaire.
The codes in the cover page of accomplished questionnaires can be checked against the list of geographic identification of sample areas (for surveys or administrative-based data collection) or all areas (for censuses). If the codes are not yet written on the cover page of the questionnaire, the processors should translate the geographic identification into codes before these are encoded into the computer.
Checking for consistency and completeness
All accomplished questionnaires should be reviewed according to the instructions laid out in the processing instructions. In general, questionnaires are inspected to determine if all the required items have entries, if they are consistent with each other, and if the values are reasonable.
It is also necessary that edited questionnaires be subjected to verification by the supervisor on a sample basis (approximately 20% of the total edited questionnaires). This procedure ensures that missed or overlooked items during the first phase of editing can still be corrected by the supervisor. Moreover, this procedure determines if the editor's corrections in the questionnaires are reasonable and are in accordance with the instructions provided in the processing manual. If there is any error in the editor's correction, the supervisor should call this to the editor's attention so that the same error will not be repeated in other questionnaires.
After the questionnaires have been edited, they are ready for the information to be transferred into the computer program. There are two common methods of capturing the data - data entry and scanning.
This is done through the process of encoding or "keying-in" of the information from the questionnaires into the computers. This is sometimes called "traditional data entry".
Design of the data entry program
In the design of the data entry system or program, some checks on the maximum or minimum values (range checks) can be included so that even during the data entry stage the validity of the data is protected.
One way to guarantee the accuracy of the coding is to re-key-in the values and then match the keyed values during the first encoding to the values during the second encoding. If there are inconsistencies, it means that some values have been mis-keyed. In this case, the verifier should determine which values are correct by checking them with the original documents/questionnaires. It is not necessary, however, that all questionnaires be keyed-in twice. The sampling rate will depend on the degree of errors in the first data encoding stage.
Another process of data capture is scanning. In this process, the questionnaires are fed into a copier-like machine where the image of the completed questionnaire can be viewed from the screen. Some of the common technologies in this method are:
- Intelligent Character Recognition (ICR) - this is used for hand-printed data recognition.
- Optical Character Recognition (OCR) - this is used for machine printed data recognition.
- Optical Mark Recognition (OMR) - this is used for recognizing marks in circles, squares or ovals.
Each one has its own use depending on the format and content of the questionnaire.
The advantage of scanning over data entry is the speed of capturing the data. However, while data entry has the problem of mis-keyed values, scanning also has problems regarding recognition. The system analyst as well as the data collection developer should therefore study the advantages and disadvantages of each technology in terms of accuracy, efficiency and cost.
Geographic identification (GEO-ID) check
After the data capture, the geographic identification of each questionnaire should be checked by the computer. It is true that these have been checked already during the receipt and control and general screening process (manual processing stage) but there are always human errors that we want to avoid. The GEO-ID check is very important because if it is incorrect then the information in that questionnaire may be erroneously attributed to the wrong area. This will lead to under or over estimation of values in the areas.
GEO-ID check also identifies those questionnaires that were not keyed-in or were overlooked by the data processors. If there are errors identified in this kind of check, the computer should create output (reject listing) containing the list of errors for the processors/editors to verify, either from the questionnaire or from the list of geographic areas.
Design of GEO-ID check program
The GEO-ID codes of the keyed-in or scanned questionnaires are matched with the master file. The master file contains all the GEO-ID codes of all the expected areas. If there are any unmatched or incomplete questionnaires, the computer will display the unmatched information for verification by the processors. After verification, the corrected information is keyed into the computer for updating. After updating, the GEO-ID Check Program is again generated. The computer will display the same errors if the computer senses the same unmatched identification. The process will be repeated until all the errors are corrected.
This kind of computerised check is done after the GEO-ID Check. This determines if all the required sections or modules (for example, body function module, vision module or section) are encoded in the questionnaires.
Design of the structural check program
All the required section or module numbers (or items in the questionnaire) are searched by the computer. If any missing sections modules or items are detected, the computer displays these and the identification of the questionnaire so that the editor/processor can check with the actual questionnaire. Once the error is found, the correction is encoded in the computer and the same structural check program is run. This process is repeated until all errors are corrected.
Consistency edit check
After the data has passed the structural check, the next stage in computer processing is the consistency edit check. Even if there are instructions during manual processing on how to edit and verify the items in the questionnaire, there is still a need to do the same in the computer due to the possibilities of mis-keying data, incorrect editing and verification by processors.
Design of consistency edit check
In this stage, the relationship of one item to another is checked. The computer will identify a case as an error if the relationship does not agree with the relationship specified as correct by the statisticians. For instance, age is checked against the highest educational attainment of individuals. The computer will detect it as an error if, for example, the age is 10 years old but the education reported is college graduate. In this case, the computer displays the possible sources of error and the corresponding identification of that questionnaire for the processor to verify with the actual questionnaire. The database should be updated once the correct values are known.
Several relationships between variables can be included in this program. Like in the GEO-ID and structural checks, this process should be repeated until all the errors have been corrected.
In some instances, especially if there are few sections or modules in the questionnaire, the structural check and the consistency check can be combined in one program so as to minimize the number of times that the checking procedure needs to take place.
In some cases, it is not possible to verify the data with the questionnaires, especially if there are a very large number of questionnaires, as in the case of a census. If this is the case corrections can be made using the output (or reject listing) generated by the computer. The output should include not only the error of the data but also the values of related items in the questionnaire so that the correct value can be determined from the values of related items in the questionnaire.
In this stage values are assigned to any remaining missing entries. These are assigned automatically by the computer program, on the basis of inputs provided by statisticians or subject matter specialists. An example is as follows:
There is no reported age of a female respondent, but the age of her husband is reported. In this case, the subject matter specialist can ask the programmer to assign an age for the spouse in the imputation program. This is done on the basis of known facts about the relationship of the ages between husbands and wives in that country. For instance, if the average age of husbands is two years greater than their wives', then the computer can assign the age of the female respondent to be the age of her husband minus two.
Design of Imputation Program
The design of the program depends on the kind of imputation procedure to be followed. One approach is to rely on known information, as in the case of the example given above - ages of husband and wife (this is called "cold deck approach"). Another approach is to use the valid information gathered from other questionnaires ("hot deck approach").
Generation of Preliminary Tables
The last check in computer processing is the preparation of preliminary tables. It is advisable that the tables be generated at the lowest geographical subdivision, e.g, by province, so that the distribution can be easily studied and outliers can be easily traced. If there are outliers, a computer program can be developed to trace the causes of such outliers. As in the previous checks, this stage will be repeated until there are no more outliers in the tables.
Generation of Final Tables
Generation of final tables should be done only after the data has passed through rigid checking procedures. This ensures that the data provided to users is of high quality.