Analysis guidelines for Quantitative Research General Guidelines Correct any data errors. As a result of initial descriptives you may detect errors (such as Sex of Subject with values of 1, 2 & 3; or a 20 year old with a child that is 17 years old). If you carefully study the output of your descriptive analyses and correct the errors now, it will lessen the chance that you will have to rerun dozens of analyses later on after discovering data errors. Use meaningful variable names such as AGE, NUMSIBS; don't use VAR01, X3, and the like. Recode topic variables. In general recode all variables so that a higher score means 'better,' 'more positive,' 'more desirable'. Form indices. After you have recoded as above, compute index variables, such as WORK = (COWORKER + BOSS + PAY)/3.0. Do later analyses on the raw scores and also on these index variables. Recode demographic variables. In most cases, your demographics should be recoded to have 2 or 3 levels at most. Let's say you have say measured Single (25), Married (15), Remarried (0), Widowed (2), Divorced (6) for Marital Status, where the numbers in parentheses are the cell counts. Look at the cell counts for the conditions and recode into two groups that will result in the larger cell counts per condition; in this case recode into Single (25) and Married (23) where every condition but Single has been recoded into Married (and hence now means either currently married or at one time married).
Example showing means, n, percentage From: Violence, Conflict, Trickery, and Other Story Themes in TV Ads for Food for Children (Rajecki, McTavish, Rasmussen, Schreuders, Byers, Jessup) Abstract Stories in a sample of 92 TV food ads aimed at children were analyzed for thematic and subtextual content. Violence as a surface theme ranked first in use receiving a nonzero score in 62% of the ads, followed by conflict (41%), achievement (24%), mood alteration (23%), enablement (18%), trickery (20%), and product dependence (8%). Cluster analysis identified six groupings of themes, with 64% of the ads characterized by some combination of violence, conflict, and trickery. Regarding subtexts, the computer-based Minnesota Contextual Content Analysis program evaluated the voiced material in ads in terms of four marker categories named "traditional (normative)," "practical," "emotional," and "analytic." It was found that the texts had a strong emphasis on the "emotional" subtext or thrust, and a pronounced underemphasis on the "analytic" context. These analyses identify possibly dubious content in a significant segment of children's TV viewing. Results Table 1 shows average theme scores for the 92 ads that offered a story, presented separately by the genres of live, animated, (or mixed characters--not shown). Overall weighted means for themes are also presented, as are the percentages of ads (rate) receiving a nonzero score for a given theme. It can be seen that the themes of violence and conflict were frequently encountered in our sample of children's food ads. The remaining themes were also widely represented, but at lesser rates. Table 1 Average Scores for Story Themes Per Genre, Overall Weighted Average Scores (M), and Percentages of Ads (Rate) Receiving a Nonzero Score for a Given Theme _________________________________________________________ Per genre Overall ___________________ __________________ Theme Live Animated M Rate (n = 25) (n = 42) (n = 92) _________________________________________________________ Violence 0.52 1.19 0.87 62% Conflict 0.52 1.17 0.75 41% Achievement 0.80 0.12 0.36 24% Mood alter 0.36 0.36 0.36 23% Enablement 0.08 0.43 0.35 18% Trickery 0.20 0.38 0.28 20% Dependence 0.48 0.02 0.14 8% _________________________________________________________ Note. Theme scores ranged from 0 to 2, and themes are ordered in terms of descending average scores.
Example showing correlations From: Lay Judgments of Thinking and Feeling in a Dog: An Implicit Anthropomorphic Model of Canine Mentality (Rasmussen & Rajecki) Abstract A lay model of canine mentality was approximated by comparisons with lay estimates of human child mentality. Independent groups of undergraduates rated capacities of a dog or a boy on 12 categories of thinking and 30 items of remorseful feelings for misbehavior. The boy received superior ratings for so-called "complex" (but not "simple") thinking categories and "upper level" (but not "lower level") remorse items. Even so, there were associations between dog and boy means across all 12 thinking categories (r = .74) and all 30 remorse items (r = .72). These several comparisons were taken to indicate that whereas the dog and boy were perceived as having mentalities that were quantitatively different, those respective mentalities were nevertheless viewed as qualitatively similar. Results Three sets of analyses were carried out: one in which the combined dog and boy data were employed, and two in which the dog and the boy were treated separately. For all analyses separate correlations were calculated between the two MO indices and the two G/S indices; these are given in Table 4. ----------------------------- Insert Table 4 about here. ----------------------------- As the top (dog and boy combined) panel in Table 4 shows, individuals' scores on both the simple and complex MO composites were moderately predictive (.28 and .36) of their positions on the lower level G/S index. That is, respondents who assigned relatively high simple or complex thinking scores were somewhat more likely to perceive the transgressing actors as experiencing guilt or shame of the lower level variety. In contrast, respondents who assigned relatively high complex thinking scores were quite likely (.56) to also assign relatively high scores on the measures that fell in the upper level G/S category. Of equal import, position on the simple MO composite was effectively unpredictive (.13) of standing on the upper level index. That is, for the combined dog/boy analysis perceptions of upper level guilt and shame were related only to mental ascriptions of the complex sort. Mental operation categories that loaded on Factor 2 (in Table 2) were those that accounted for variation in the upper level G/S data set. Roughly the same pattern of coefficients emerged when the separate dog and boy Pearson correlation analyses were carried out. A notable similarity across the three panels in Table 4 is the relatively weak predictive power of the simple thinking MO variable where the upper level G/S composite was concerned. These particular cell entries (.13, .11, and .22) indicate that the first variable accounted for only 1% to 5% of the variance in the second. The association of complex thinking and the upper level G/S composite was stronger with variance accounted for ranging from 11% to 31%. Thus knowing how a subject stood on the complex thinking composite (but not the simple composite) distinguished to some extent her or his standing on the G/S upper level composite index that identified the dog and boy as differentially remorseful for their transgressions. Table 4 Pearson Correlations (r) between MO Factor Composite Scores and G/S Derived Composite Scores for Both Actors Combined, and Separately for the Dog and Boy Actor Types ___________________________________________________________ G/S derived composite _______________________________ MO factor composite Lower level Upper level ___________________________________________________________ Dog and boy actor types Simple .28*** .13 Complex .36*** .56*** ___________________________________________________________ Dog actor type Simple .30** .11 Complex .28** .33** ___________________________________________________________ Boy actor type Simple .30** .22* Complex .43*** .48*** ___________________________________________________________ *p < .05. **p < .01. ***p < .001
Example showing means and F ratios From: Lay Judgments of Thinking and Feeling in a Dog: An Implicit Anthropomorphic Model of Canine Mentality (Rasmussen & Rajecki). Abstract A lay model of canine mentality was approximated by comparisons with lay estimates of human child mentality. Independent groups of undergraduates rated capacities of a dog or a boy on 12 categories of thinking and 30 items of remorseful feelings for misbehavior. The boy received superior ratings for so-called "complex" (but not "simple") thinking categories and "upper level" (but not "lower level") remorse items. Even so, there were associations between dog and boy means across all 12 thinking categories (r = .74) and all 30 remorse items (r = .72). These several comparisons were taken to indicate that whereas the dog and boy were perceived as having mentalities that were quantitatively different, those respective mentalities were nevertheless viewed as qualitatively similar. Results Importantly, there was also a significant Actor Type X G/S Item interaction, F(29,4988) = 3.32, p < .001, which is shown in Table 3. The table gives the averages for the ----------------------------- Insert Table 3 about here. ----------------------------- G/S items for the dog and boy actor types, rank ordered in terms of F ratios contrasting individual pairs of means. The F ratios, which were calculated using separate error variance for each item, ranged from F(1,186) = 39.14 to 8.03 for the significant ratios (p < .01), and F(1,186) = 7.36 to 0.00 for the nonsignificant ratios (p > .01). (Given the number of comparisons carried out, it was deemed more appropriate to use the p = .01 than the p = .05 level to establish significance.) By convention, entries in the top panel of Table 3 are termed "upper level" items, and those in the bottom panel are termed "lower level" items. Given the unproductive factor analysis of G/S material (above), this upper/lower distinction provides at least one meaningful way to sort out subsets of these items. Of course, by choosing the terms "upper" and "lower" we mean to connote a qualitative distinction between the two levels, just as in the case of simple and complex thinking from Rasmussen et al., 1993. Table 3 Mean "Reasonable to Say" Scores for G/S Items, and F Ratios for Individual Items Defining Upper and Lower Level Sets ______________________________________________________________ Actor type ____________ G/S item Dog Boy F ______________________________________________________________ Upper level items feel embarrassed 3.38 5.01 39.14 try to distract self 3.69 5.29 36.84 think back a lot 2.98 4.50 34.96 ...additional items... ______________________________________________________________ Lower level items feel naughty 4.12 4.85 7.36 feel sad 4.03 4.67 6.18 feel alone 4.04 4.63 4.76 ...additional items... ______________________________________________________________ Overall (N = 30) M 4.06 4.81 Upper level items (n = 16) M 3.69 4.79 Lower level items (n = 14) M 4.47 4.82 ______________________________________________________________ Note. Individual items are ordered in descending order of F ratios.
Example showing factor analysis From: People's Perceptions of Animal Mentality: Ascriptions of "Thinking" (Rasmussen, Rajecki & Craft) Abstract On standardized rating scales student respondents indicated whether it was reasonable to say that a dog, cat, bird, fish, and school-age child had the capacity for a dozen commonplace human mental operations or experiences. Factor analysis of responses identified two levels of attributions: "simple thinking" and "complex thinking." The child, and to a lesser extent all the animals, were credited with simple thinking, but respondents were much more likely to ascribe complex thinking to the child. (A small pilot study involving animal-behavior professionals generally replicated these results.) Certain mental categories (e.g., emotion) were judged by students to be simple for all target types; others (e.g., conservation) were judged to be universally complex. However, further factoring revealed articulate ascriptions for key mental categories: play/imagine was seen as simple in the animals, but complex for the child. Contrarily, enumeration/sorting and dream were seen as simple in the child, but complex for the animals. The results were taken as an addition to a lay theory of animal mind. Results Given the large number of categories, it was desirable to reduce the original list to a smaller set of hypothetical factors. To this end factor analyses were conducted for each target type separately. Kaiser's "eigenvalue greater than one" criterion was used to aid in the selection of the maximum number of factors. A two-factor varimax orthogonal solution was thus employed for each of the analyses (the BMDP 4M program was utilized; see Dixon, 1990). A visual inspection of the factor loadings indicated that the factor structures for the five target types were highly similar to one another: 90% of the categories loaded on the same factor across the five analyses. Furthermore the factor loadings for Factor 1 for the five solutions were highly intercorrelated with one another, as were the factor loadings for Factor 2. The average correlation of the factor loadings was r = .83, with a range from .60 to .99 (all significant at p < .05). Because of the similarity of the factor solutions, another varimax orthogonal factor analysis was run on the 12 categories, but ignoring target type. Again, Kaiser's criterion was used to aid in the selection of the number of factors. The two-factor solution is given in Table 2. --------------------------- Insert Table 2 about here --------------------------- Entries in the table are Pearson's correlations of the scores from the observed dimensions with the hypothetical factors generated by the factor analysis program. As Table 2 shows, Factor 1 is concerned with relatively simple mental operations or experiences, whereas Factor 2 is concerned with more complex categories. The first factor accounted for 37.03% of the variability of the dimensions, and the second accounted for 25.71%. To reduce the 12 dimensions into two indices the average of the unweighted linear composite of the first five categories (Factor 1) listed in Table 2 was formed into one index, and that from the remaining seven categories (Factor 2) was formed into another. Taken together these make up two levels of a repeated measures variable. That variable will be referred to as "factor." Given that the respondents had been given instructions about thinking, the two levels of this variable will be arbitrarily called "simple thinking" versus "complex thinking." Table 2 Common Factor Loadings over All Target Types for Students _________________________________________________________ Factor ______________________ Category label 1 2 _________________________________________________________ Sensation/perception .79 Pleasure/displeasure .78 Emotion .72 Gratitude .68 Play/imagine .59 Conservation .84 Enumeration/sorting .78 Memory/foresight .77 Schemata .76 Morality .76 Dream .69 Object permanence .65 _________________________________________________________ Note. Loadings less than .50 are not shown. Category labels are listed in descending order of factor loadings.