STANDARD MULTI-SURVEY SHELLS IN NASS

Transcription

1 STANDARD MULTI-SURVEY SHELLS IN NASS Roger Schou and Mark Perzchala Natonal Agrcultural Statstcs Servce, USA 1. Background A "shell" refers to standard Blase programs contanng begnnng and endng admnstratve modules that surround survey-specfc questons. Three related shells wll be used n almost all surveys conducted n Blase n the Natonal Agrcultural Statstcs Servce (NASS). These are the Lst Frame Shell, the Area Frame Shell, and the Multple Frame Shell. The Lst Frame Shell and the Multple Frame Shell are now complete. The Area Frame Shell wll be adapted from the Multple Frame Shell by early In order to understand why these shells are mportant to NASS and why they are so complex, t s necessary to know a few thngs about the samplng schemes of NASS and the basc survey operatons ncludng nteractve survey processng on NASS's many Local Area Networks (LANs). NASS has 45 state offces, headquarters n Washngton, DC, and a research faclty n Farfax, Vrgna. Ths means that there are potentally 47 programmng stes and 45 end-user stes for varous natonal-level and statelevel Blase applcatons. Regardless of ste, NASS employs standard samplng frames, samplng schemes, and survey procedures. Ths ensures that data collected n each state can be aggregated to natonal-level estmates. The shells embody standard NASS methodology for almost all surveys. By developng and dstrbutng the shells throughout the agency, programmers can concentrate on just the survey-specfc questons thus avodng much duplcaton of effort. The shells ensure that surveys are conducted n approved ways wthout unduly lmtng flexblty n each state offce. The complexty of the shells derves n part from the use of two conceptually dfferent sample frames, the Lst Frame and the Area Frame. The samplng schemes based on the two frames requre that sample weghts take Second Internatonal Blase Users Conference (1993) 133

2 nto consderaton not only samplng rates wthn strata but also the degree of overlap between the lst and area frames and of the duplcaton of potental respondents n the lst frame. These consderatons requre many detaled and related questons about each ndvdual assocated wth the farm or pece of land n order to determne who operates t. Farmng operatons can have many dfferent types of ownershp and operatng arrangements ncludng ndvdual ownershp, partnershp, and corporate management. Another major contrbuton to the complexty of the shells stems from the many tasks the Blase system s beng asked to do from one source code ncludng nteractve edtng, Computer Asssted Telephone Intervewng (CATI), Computer Asssted Personal Intervewng (CAPI), survey management ncludng call schedulng, and some data entry from paper forms. The fact that Blase wll handle multple tasks s further complcated by the mult-mode nature of data collecton n NASS. For most of the man agrcultural surveys, data are collected both on paper and by computer asssted methods, prmarly CATI. However, NASS wll soon start a plot program where data are also collected on laptop ether n person (CAPI) or by phone from feld enumerators' phones (decentralzed CATI).When data are entered from paper forms, NASS presently uses a hgh-speed data entry package called Key Entry 3 employng tem code data entry. Ths means that only data 'wth tem codes are keyed. In NASS questonnares, many of the admnstratve and screenng questons do not have tem codes, yet they are essental to the flow of both the paper and the computer asssted ntervew. However, they are not needed for summarzaton and not keyed when the data arrve on paper. Snce these questons are not keyed, much of the data from paper forms would be off the route f drectly entered nto an nstrument that s constructed solely for ntervewng. Therefore, the shells must take nto account the source of the data and adjust the route accordngly. There are other major dfferences between computer asssted ntervewng and nteractve edtng n NASS. For example, an ntervewer may be allowed to enter a DontKnow or Refusal for an tem durng an ntervew, but ths s somethng that must be called to the attenton of the data edtor n order to personally mpute a plausble value. Some edts may be soft n the 134 Second Internatonal Blase Users Conference (1993)

3 ntervew but hard n edtng. It s NASS polcy to requre each approprate queston to be asked n computer asssted ntervewng requrng an answer of at least zero for many questons. On the other hand, durng hgh-speed tem code data entry, zeroes are not entered. Thus, a way must be provded to allow empty answers n edtng whle requrng at least zero answers n ntervewng. NASS makes maxmum use of Blase's processng modes, CADI for edtng, CATI and CAPI for ntervewng, and the ablty to condton routes and edts on these modes. NASS also needs to condton routes and edts on survey task, for example data entry or verfcaton. Table 1. Processng scenaros Processng scenaro KE3 - IE BDE - IE CAI - IE CAI - CAI - IE CAI - IE - CAI - IE IE - CAI - IE CATI - CAPI - IE Remarks Data questons only on the route. Update call scheduler and NASS survey management system. Data questons only on the route. Update call scheduler and NASS survey management system. Keep all modules on route but nvoke IE-specfc tasks when approprate. Update call scheduler and NASS survey management system. Update call scheduler and NASS survey management system when an ntervew s nterrupted. Keep all modules on route but don't undo any IE-specfc work when gong back to CAI. Update call scheduler and NASS survey management system. Put all CAI modules on route when necessary. Update call scheduler and NASS survey management system. Update call scheduler, NASS survey management system, and CAPI management system. Second Internatonal Blase Users Conference (1993) 135

4 The shells provde hooks nto NASS's other major data processng systems, for example wth the samplng frames and a survey management system. In order to allow maxmum flexblty for each state offce, the shells must be able to handle each form ndvdually n each of the scenaros n table 1. In the table, CAI = Computer Asssted Intervewng, IE = Interactve Edt, KE3 = Data Entry n Key Entry 3, and BDE = Blase Data Entry. 2. Desgn of the Multple Frame Shell The shells have been developed to accommodate the specfcatons that have been generally explaned above. They have been bult n modules. The modular approach was adopted n the Multple Frame Shell n order to facltate mantenance, modfcaton to other shells, and as a bass for the subfle layout. The Multple Frame Shell and the Lst Frame Shell are related n that they both are bult of many of the same modules. Generally, the Lst Frame Shell.uses a subset of the Multple Frame Shell's modules and altered versons of some of the others. Thus f a common module must be modfed, t has to be modfed n only one place. The modules are stored n a formal lbrary of Blase Code-that s beng bult up to facltate the producton of nstruments. Fgure 1 presents the general structure of the Multple Frame Shell, when each module s nvoked for a few of the processng scenaros above, and where data are stored. Extensve use s made of the Blase subfle faclty n order to ease pre- and post-survey processng of admnstratve data. For example, name and address updates appear n subfle 07 and do not have to be separated later from the data to be summarzed. In addton, each subfle can be processed on ts own schedule because Blase keeps status nformaton for each subfle separately. For example f data are read out from several subfles but then data are changed n one subfle only, then t s possble to read out only the data that have been changed, by subfle. The use of subfles also keeps data from beng nadvertently overwrtten. Durng ntalzaton, certan control data are stored n subfle 03. When survey data are then read nto the man data fle, the control data wll not be overwrtten wth blank values as 136 Second Internatonal Blase Users Conference (1993)

5 Fgure 1: Schematc of Multple Frame Shell When nvoked Modules ll Data storage 1 2 r, 345 ',. T t 1 T T Id Nonresponse Appontment Control Felds Swtches Management Current Operator Master Operator CAPI Info Screen Introducton Verfcaton Screen Operator Screen For Survey Change In Operator Screen Operaton Partnershps SURVEY QUESTIONS Another Operaton Operator Numbers Offce Use Man F le [( ~ 1 <n Subfl [of] fotj 1 Data are read n durng ntalzaton 2 Data are computed durng ntalzaton 3 Invoked n CAI IE scenaros 4 Invoked n IE - only scenaros 5 Computed durng data fle movement 11 See Table 2 for an explanaton of the modules. Second Internatonal Blase Users Conference (1993) 137

6 they would be f they were also stored n the man data fle. Fnally, the use of subfles saves a great deal of dsk space. For example, the partnershp nformaton nvolves numerous questons. Snce these are stored n a subfle, no storage space s taken up for farms that do not have partners. Only a small percentage of farms do have partners. In table 2 below, s a lstng of each major module n the Multple Frame Shell and ts functon. Table 2. Functon of each module Module Id Nonresponse Appontment Control Felds Swtches Management Current Operator CAPI Info Screen Functon Identfcaton and admnstratve questons as well as questons used to flag forms for specal handlng. Survey-specfc module contanng global questons asked as a last resort n the event of a refusal. Automatcally codes some admnstratve questons. Standard Blase appontment block plus NASS addtons that allow CATI/CAPI compatblty and hooks nto MASS'S other survey admnstratve systems such as the Name and Address fle and the operator comment fle. Holdng place for up to 10 sample control varables as mght be requred n a partcular survey. Processng swtches (and selectors) for KE3 data entry, forms check n, and the Blase utlty Forms Manager. NASS-specfc survey management questons are set here n order to defne survey management reports once that apply to all surveys. A Pascal sub-routne reads DOS envronment varables for the user of the system. Name and address of the person currently consdered to be the operator. Master OperatorName and address of the person thought to be the operator at the tme of ntalzaton. Dsplays comments about the operator/operaton f the CAPI mode s used (and hence the CATI call scheduler s not used). 138 Second Internatonal Blase Users Conference (1993)

7 Standard Mult-Survey, Shells n NASS Introducton Verfcaton Screen Operator Screen for Survey Change n Operator Screen Operaton Partnershps Another Operaton Operator Numbers Offce Use Introductory questons for ntervewers, determnes respondent, hs/ her relatonshp to the operator, and follow-up questons for outof-busness operatons. Name and address verfcaton and updates for the operator and operaton.. Checks to see f any operators of an operaton have been added or dropped. Survey-specfc screenng for commodtes of nterest. Invoked when an operator s out of busness. Tres to fnd the new operator and decdes whether a new form s needed for substtuton based on screenng.... Determnes type of operaton (ndvdual, partnershp, managed), and asks follow-up questons f necessary. Name and address verfcaton of up to 6 partners, determnes whch partner s the operator. Allows addton and deleton. of partners. Determnes f the operator s responsble for another operaton. Gathers senstve operator/operaton numbers such as Socal Securty Number and Employer Identfcaton Number. Admnstratve data entered only n the clercal or statstcan edt. 3. Other aspects of the Multple Frame Shell The complexty of the Multple Frame Shell s very dffcult to convey n ths short artcle; however, one ndcaton of ths s ts sze. When compled wthout any addtonal survey-specfc questons the shell contans-428 unquely defned and 745 overall used questons n addton to several hundred computatons.and edts. For many of. NASS's surveys, the number of questons n the shell wll outnumber the number of survey-specfc questons. Many of the modules are ntrcately nterconnected. For example, when a change n operator-s dscovered approprate changes-have to be reflected n Second-Internatonal Blase. Users. Conference (1-993) 139

8 several dfferent subfles. Ths way the queston flls for the ntervewer wll always be up-to-date and the call scheduler wll brng the corrected names and addresses onto the screen n the case of an nterrupted ntervew. The shell s also desgned to be very robust. Ths means that consderable change can be made durng or between surveys wthout dsruptng callng or edtng. In addton to the Blase code, a "protocol" around each shell has also been developed, whch s part of any NASS nstrument. In NASS, the term protocol refers to the subsdary programs whch surround survey processng. These nclude programs whch transfer names and addresses from the frames nto Blase and back agan f there are changes, electronc movement of forms from one Blase data set to another, generaton of survey management reports, the formattng of external fles holdng hstorc data, etc. Almost all of the protocol has been developed n Blase utltes, especally Manpula (fle manpulaton and reports), Abacus (tabulaton), and Forms Manager (movement between Blase data sets). 4. Development experence and mprovements n Blase The development of the shells has proved to be an teratve experence as NASS works to polsh. procedures for nteractve processng wth multple modes of data collecton usng one source code and the same data sets. One of the most dffcult aspects of development has been to fnd out exactly what needs to be programmed. NASS s just movng to a new screenng algorthm whch whle smpler than the old one, stll requres qute detaled programmng of many dfferent rules whle accountng for many exceptons to the rules. In order to learn all of the procedures, many people were ntervewed and the shells have passed through many revews. An early verson of the Lst Frame Shell has been n operaton for two quarters n the Catfsh survey. The 15 state offces that conduct ths survey have provded valuable feedback on the performance of the shell. Ths shell has performed adequately n ths survey wth one major omsson beng that 140 Second Internatonal Blase Users Conference (1993)

9 Standard Mult-Survey Shells: n MASS when data were read n from Key Entry 3 the call scheduler was not. updated, properly. Thus sometmes a call would be placed to someone :who had already sent n a questonnare. Ths omsson has.been :taken care- of n the current versons of the shells ,. -., The Multple Frame Shell was modfed n a few days to serve as an ad hoc Area Frame Shell for the 1993 June Area Frame Survey research project. Some code was commented out and a few modules were added, but yet t was far easer to modfy the already exstng shell than to start from scratch. The changes that were made took nto account the conceptual dfferences between the Lst and Area Frame samplng as done n NASS; however, the majorty of the code that was kept ensured that the mechancs of screenng (.e., partnershp verfcaton) was done the same way between the two types of surveys. The Area Frame Shell wll be formalzed and more rgorously tested n early 1994 based on the experence ganed n 1993.A few of the dffcultes n programmng the shells can be traced to the Blase language and the Blase way of dong thngs. As a result of the June Area Frame project, NASS dentfed 7 specfc language enhancements that would ease mult-mode and mult-task nstrument development. These suggestons have been forwarded to the Netherlands CBS and t appears that most or all of the suggestons wll be mplemented n Blase, verson III. 5. September Quarterly Agrcultural Survey The frst operatonal test of the Multple Frame Shell wll have been conducted n September 1993 n three state offces. A companon paper by Asa Mannng descrbes the plannng that has gone nto ths tral. NASS fully expects to revse both the Multple Frame Shell and the offce procedures as a result of the September tral and then to test the revsons n December n the same state offces. The development of a processng flow n the mult-mode and mult-task envronment has been tself teratve. Four dfferent flows have been devsed, each of whch would work, but each havng strengths and weaknesses. The fourth s a combnaton of two earler flows. One of the bggest concerns has been the potental for nterrupton of some Second Internatonal Blase Users Conference (1993) 141

10 processes n order to perform other tasks. For example, f ntervewers and edtors are workng off of the same data set, then the CATI would, have to stop n order to perform a batch edt. In consultaton wth the Netherlands CBS, NASS has devsed a scenaro that should much reduce or elmnate tasks nterferng wth each other Second Internatonal Blase Users Conference (1993)