SN: 3332 FAMILY RESOURCES SURVEY 1993/94 Note on data quality by the Department of Social Security Background The FRS began in October 1992. Following a six month period largely designed to test the fieldwork operations, the first complete year began in April 1993. Each year, the questionnaire is reviewed. As well as new items being added, questions which have not worked so well have been highlighted. Where appropriate, these have then been revised or additional interviewers' notes added. Under the terms of the first contract (which ran until March 1997) responsibility for provision of the data base to users lay largely with the DSS. Although some office editing was done by the fieldwork contractors, the bulk of validation and editing not included as part of the interview, together with imputation and creation of derived variables, was carried out by the Department. More information on what this actually involved is given in the annual FRS report. Data were provided by the contractors in a format which had to be converted back into a data base structure. This is a complicated and time consuming task, which involves recreating the structures of the original Blaise interview. Over the lifetime of the contract, the Department's understanding of this process has grown and methods of conversion and associated checking have improved. At the same time, methods of imputation and other tasks have similarly developed. Implications The FRS is still a relatively new survey for which the questionnaire and procedures have yet to reach a steady state. Each year changes and improvements have been made. This means that the latest year which is available (currently 1995-96) will provide the best quality information. However, changes and improvements affect only the minority of cases: for example, improvements in imputation only affect the 0.5 per cent of values which are missing. Earlier years' data will still be usable for the vast majority of purposes. FRS 1993-94 At the User Group meeting held last October, a particular issue was raised relating to Social Security benefits data for 1993-94. This was the potential double counting of benefit income for cases where more than one benefit had been paid in combination. This arose as a result of the way interviewers were instructed to record such payments. Where benefits were paid together, they were instructed to put the total combined amount under one benefit and leave the other(s) as "don't know". Cases were then put through the imputation system and all benefit "don't knows" imputed, ie double counting the amount included in the recorded (total combined) amount. The spreadsheet benmis30.xls lists all cases on the 1993-94 survey where more than one benefit was recorded and at least one was missing. This is the pool of potential combined cases where income may have been double counted. It may be appropriate to make an adjustment for these cases in some analyses. This problem does not exist in any other year. Validation and imputation of the benefits questions has been progressively improved. The future The Department is currently reviewing the conversion of, and some of the imputation methods used, on the 1994-95 data set. This is likely to result in a new version of the data set being deposited with the Data Archive later this year. There are no plans to redeposit the 1993-94 data set, although an unimputed data set will be created. If there was a demand from users for this data set, it could also be deposited. The FRS questionnaire, methods of imputation and derived variable creation will continue to be improved. From April 1997 responsibility for creation of a data base has been given to the fieldwork contractors. Their knowledge of the structures, developed in writing the Blaise instrument, should minimise conversion errors in future years. They have also taken on much of the editing and validation, based on rules supplied by the Department. All other processing remains the responsibility of DSS. Extensive checking of the conversion process for 1995-96 was undertaken prior to deposit at the Data Archive. Similar checking of 1996-97 will be undertaken before it is finally ready for release. 1995-96 data are thought to correctly reflect the original FRS interview. However, as the data are used more widely, problems affecting a tiny minority of cases may yet be revealed. Should this happen, the Archive will be informed and new files released. High values found on the FRS version 30 database by the ESRC Essex Archive: Summary Several cases on the 1993/94 data base have been identified by the ESRC Archive as having high amounts. In some cases relating to income and expenditure, this is where figures have not been converted to weekly amounts, because period codes for the payment are equal to 12 (one off payments) or 13 (none of the above). Other cases may be due to interviewer error and it is up to users to decide whether data should be amended. Only limited editing was carried out by DSS prior to finalisation of the data base. HBENRAMT Reduced figure for rent: high value - £2869 Case 7083051 - rent paid was £28.74 per week. Suggest decimal point may be misplaced and the value should have been £28.69 ACCAMT Rent paid by someone outside the household: high value - £1500 Cases 1913021 and 1913022 - Both these cases had period code 13 ("none of the above"). The figures quoted having not been converted to a weekly figure (see file dssedit.v30 for more details). SERVAMT Service charges: high value - £105 Case 5983181 - two bedroom accommodation, with the services for cleaning and gardening, amount may have been incorrectly input. PURCAMT Cost of house purchase: high value - £38500383 Case 13173071 - 4 bedroom accommodation, the mortgage borrowed was £20,000, top up £8,000 and the estimated value was £80,000. Suggest data incorrectly input. ESTWORTH Estimated value of property: high value - £2200000 Case 3333231 - 4 bedrooms, mortgage borrowed was £30,000, no top up and the purchase price was £65,000 in 1982, possible increase due to rising house prices but this much unlikely. Suggest figure has been incorrectly input, although respondents' ability to estimate the value of their property has been questioned. TOPEX Value of re-mortgages and top ups: high values - £270000, £220000, £200000 Cases 7143271, 15733011, 11843161 - All of these cases had the same characteristics, that of 4 to 5 bedrooms, with the original purchase price very low, but the estimated value now quite high £340,000, £200,000 and £140,000. Two of the cases were self-employed couples with the third case having a high gross wage of £997.23 pw. These could safely be genuine cases, especially with the self-employed couples, where individuals have re-mortgaged there property to release capital. OUTSAMT Amount paid by someone outside household towards the mortgage: high values £15000 Case 12293061 - period code 13, not converted to a weekly amount CCAMT Community Charge paid high value - £500 Cases 13293251, 7033181 - both cases had period code 12 or 13. MNTAMT Maintenance received: high value - £60000 Case 12553101 - period code 12. OTAMT Amount of other tax payments made in last 12 months: high value - £3846 Case 8853041 - no evidence to suggest why such a high value, no period codes, person who paid the tax was not working but total capital was over £20,000. A final query related to INTDATE (date of interview), where some cases are shown to have been interviewed as early as 1903. This variable is input by the interviewer and does not undergo any checking. 1903 most likely related to 1993.