Site Home > Symmetrical Archaeology > Reconfiguring the Archaeological Sensibility: Mediating Heritage at Teotihuacan, Mexico > Versions for "Coding"
Coding
Coding is a key intermediate step in getting from an individual’s response to a number that can be summated with the responses to other questions. The complete codebook is included as Appendix 3. It provides the concise information deemed compulsory in any statistical study so that the study may be replicated by detailing the logical linkages and operations performed on the data by myself in moving from handwritten responses on the questionnaires to the final statistical database. It provides: 1) the exact wording of the question; 2) the abbreviated name by which the question is referred to in the statistical program; 3) the column number in the statistical program (or database) where the question’s responses are located; 4) the valid codes for all of the responses to each question; 5) the missing data codes (if any used) of each question; 6) any special coding instructions which were used in coding a question (such as the codes for ‘open format’ responses). The essence of coding is to provide a distinctive code to each answer (‘category’) of a question (‘variable’) (De Vaus 2002:148; Shennan 1988:8-12). In archaeology, such coding normally consists of quantitative measurements of the attributes of, say, ceramic sherds or of their provenance in an excavation unit (Shennan 1988:ch.2). While these seem to provide ready-made numerical codes (e.g. distinguishing between 7mm and 9mm sherd thicknesses), the codes must be salient to the goal of the research, and even here it is critical to avoid logical inconsistencies (e.g. how does one assign ‘nominal measurement’ codes to pottery decorations or designs styles) (see Shennan 1988).
In social research such as this study, the majority of responses require that codes be assigned (some categories, such as age or income, do provide ‘ready-made’ numerical codes at the ratio level, e.g. there is an unambiguous difference between a monthly income of 3,200 pesos and 11,000 pesos). For ordinal and nominal coding I used, where germane to the categories of this study, ‘industry standard’ coding schemes, such as the International Labour Statisticians International Standard Classification of Occupations (ISCO) of the United Nations for occupation codes (UN 1987). Whatever the method or source, an assignment of unique codes for every response to a question is a necessary step for any quantitative analysis to be performed.
To do this for categories unique to this study, I followed social science convention and coded all ‘affirmative responses’ with higher numbers and all ‘negative responses’ with lower numbers (De Vaus 2002:ch.9). This is, ultimately, an arbitrary decision, but was for both consistency’s sake and, for those questions that would be incorporated into scales, to make the resulting codes unidirectional and so easier to summate. This was straightforward with questions consisting of either dichotomous or multiple dichotomous format. That is, with questions resulting in either a single ‘yes/no’ response or in a series of ‘yes/no’ responses. Thus a ‘no’ response was coded as ‘1’ and a ‘yes’ response coded as ‘2’. To account for non-responses, a code of ‘999’ was entered which would not enter into calculations. For example, with question 1 (multiple non-dichotomous format) respondents were asked for their three principal reasons for being at Teotihuacan (Appendix 1, Figure 3, question 1). Eleven ‘closed format’ responses were listed, ranging from ‘to collect the spiritual energy’ to ‘learn about the scientific explanations of Teotihuacan culture’. If a particular response was selected, it was coded as 2/’yes’; those not selected were coded as 1/’no’. Additionally, the final possible response was ‘open format’, where respondents wrote in their reasons for visiting if the ‘closed format’ responses did not accurately describe their motivations. Coding these responses required content analysis, an equally interpretive practice informed by my particular research goals. This procedure entailed assigning unique codes for categories of answers which I felt to be similar in content. So the written responses of ‘to conduct ceremonies on the site” and “in order to perform rituals” were coded by me as ‘12’; the unique response “to understand world truth” was coded as ‘13’ and so on. The importance of these codes for ‘open format’ responses is not for calculation, but to make it easier in the database to find particular individuals who expressed these motivations.
Other types of question formats require analogous but expanded coding procedures. So for questions such as question 2 or question 33, asking for the number of times that the respondent has visited Teotihuacan and the education level of the respondent respectively, and where a large number and type of response are expected, I numerically coded the responses into pre-defined categories. Thus for question 1, a code of ‘1’ equals visited ‘1-2 times’, a code of ‘2’ equals ‘3-4 times’ and so forth. For question 33, a code of ‘1’ equals ‘little or no formal education’, ‘2’ equals ‘secondary school’, ‘3’ equals ‘preparatory or bachelor’s degree’ and so on. The importance of coding responses for these questions is to maintain a unidimensional order (or, in statistics, to ensure data at an ordinal level of measurement). Thus when analyzing the education level of respondents, based upon their responses to question 33, a difference between 1 and 4, while not an absolute, measurable (or ration level of measurement) quickly indicates a large difference in educational background. For questions using a Likert scale to measure attitude or opinion, this unidimensional order is easily transferred into the numeric coding. For example, all of the questions (questions 9-15, 20-23) using Likert scales were coded identically, with ‘strongly agree’ coded as ‘4’, ‘agree’ coded as ‘3’, ‘disagree’ coded as ‘2’, and ‘strongly disagree’ coded as ‘1’. The ‘neutral’ codes, or codes not calculated in statistical operations, of ‘0’ and ‘999’ were assigned to ‘don’t know/no response’ and no response (left unanswered) respectively.
The overall result of coding was to provide numbers along a unidimensional order which would facilitate analyzing particular questions (e.g. all respondents’ education level could quickly be compared by calculating the differences in the magnitudes of the numbers for that question) and be more easily combined into a scale (e.g. strong beliefs coded as ‘4’s’ to separate questions concerning archaeology). The operating assumption behind coding in this manner is that, for example, if strong agreement with questions pertaining to archaeology results in a higher score, then higher scores indicate a high association of archaeology with Teotihuacan.
The final two caveats concerning the construction of scales involve assuring that the scales have meaningful upper and lower limits, so that inter-scale comparison can be made, and, most importantly, making certain that the scales do in fact ‘measure’ or ‘tap’ the explanatory concepts or associations which I am interested in. Ameliorating these worries entails using slightly more advanced statistical operations.
Forward to Factor Analysis
Return to Explanatory Concepts
Posted at Mar 24/2009 03:57PM:
http://www.quizilla.com/stories/9060964/buy-tramadol-without-prescription buy tramadol without prescription - Click here for more information...