So, it seems the incubation effect (https://en.wikipedia.org/wiki/Incubation_%28psychology%29) really does exist. That's handy to know. "Does incubation enhance problem-solving? A meta-analytic review" http://www.psy.cmu.edu/~unsio/Sio_Ormerod_meta_analysis_incubation_PB.pdf , Sio & Ormerod; excerpts:

"A meta-analytic review is reported of empirical studies that investigate incubation effects on problem-solving. Although some researchers report increased solution rates after an incubation period, a period of time in which a problem is set aside prior to further attempts to solve, others have failed to find effects. The analysis examined contributions to incubation effect sizes of moderators such as problem type, presence of solution-relevant or misleading cues, and lengths of preparation and incubation periods. A positive incubation effect was identified, divergent thinking tasks benefiting more than linguistic and visual insight tasks from incubation. Longer preparation periods gave a greater incubation effect, while filling an incubation period with high cognitive demand tasks gave a smaller incubation effect. Surprisingly, low cognitive-demand tasks yielded a stronger incubation effect than rest during an incubation period when solving linguistic insight problems. The existence of multiple types of incubation effect provides evidence for differential invocation of knowledge-based versus strategic solution processes across different classes of problem, and suggests that the conditions under which incubation can be used as a practical technique for enhancing problem-solving must be designed with care.

Does incubation enhance problem-solving? A meta-analytic review Anecdotal reports of the intellectual discovery processes of individuals hailed as geniuses (e.g. Wallas, 1926;Woodworth & Schlosberg, 1954; Ghiselin, 1985) share a common theme: a flash of insight pops unexpectedly into the mind of the individual after they have put an unsolved problem aside for a period of time, having failed in initial attempts to solve it. This temporary shift away from an unsolved problem that allows a solution seemingly to emerge as if from no additional effort is termed an ‘incubation period’ (Wallas, 1926). Its importance in current thinking and practice is illustrated by a yield of 5510 articles that mention the term ‘Incubation’, along with one of ‘Creativity’, ‘Insight’ or ‘Problem’, from a recent search of Google ScholarTM, the search restricted to the years 1997 to 2007 and the subject areas of Social Sciences, Arts and Humanities. An additional 1970 articles were yielded by including Business, Administration and Economics. Yet there are many conflicting accounts of incubation, some studies reporting strong effects (e.g., Smith & Blankenship, 1989), others failing to find any effect at all (Olton & Johnson, 1976).

One theoretical reason for studying incubation is because it is closely associated with insightful thinking. Indeed, Wallas (1926) proposed incubation as the second of four phases in problem-solving (the others being preparation, illumination and verification). Insight may be characterized as a sudden, unpredictable and non-verbalizable solution discovery (e.g., Metcalfe & Weibe, 1987). Some researchers see the apparently unconscious nature of solution discovery as evidence that the processes required to achieve insight in problem-solving are qualitatively

Understanding the role of incubation periods may also allow us to make use of them effectively to promote creativity in areas such as individual problem-solving, classroom learning, and work environments. Educational researchers have tried to introduce incubation periods in classroom activity, and positive incubation effects in fostering students’ creativity have been reported (Lynch & Swink, 1967; Medd & Houtz, 2002; Rae, 1997; Webster, Campbell, & Jane, 2006). However, in the absence of a comprehensive theory or model that can explain how and why positive incubation effects might emerge and under what conditions they are best fostered, no general pedagogic recommendations can be made.
Several hypotheses have been proposed to account for the alleged positive effects of incubation periods on problem solving, and they can be divided into two main kinds; conscious-work and unconscious-work.
The conscious-work hypothesis holds that incubation effects are due to issues such as reduction of mental fatigue (Posner, 1973) or additional covert problem solving during the incubation period (Browne & Cruse, 1988; Posner, 1973). Both sources implicate changes in consciously controlled problem-solving activities during an incubation period. In contrast, the unconscious-work hypothesis suggests that positive incubation effects are the result of gradual and unconscious problem-solving processes that occur during an incubation period (Bower, Regehr, Balthazard, & Parker, 1990; Simon, 1966; Smith, 1995; Smith & Blankenship, 1991; Seifert, Meyer, Davidson, Patalano, & Yaniv, 1995; Yaniv & Meyer, 1987).
Three different unconscious processes have been proposed to account for incubation effects. The first involves eliciting new knowledge: over time, activation will spread towards previously-ignored but relevant memory items. Even if relevant items do not receive above-threshold activation, this process can still sensitize individuals to related concepts, and thus they will be more likely to make use of external cues to solve a problem. In addition, partially-activated concepts may combine with others to yield fortuitous insightful ideas (Bower et al., 1990; Smith, 1995; Smith & Blankenship, 1991; Yaniv & Meyer, 1987). The second hypothesis is selective forgetting: an incubation period will weaken the activation of inappropriate solution concepts that distract individuals during initial attempts, allowing a fresh view of the problem (Smith, 1995; Smith & Blankenship, 1991). The third hypothesis is problem restructuring, in which an individual’s mental representation of a problem will be re-organized into a more appropriate and stable form after initial unsuccessful attempts. The individual is then more able to capitalise upon relevant external information or to re-arrange problem information in a manner that allows a solution to be found more readily (Seifert et al., 1995). Problem restructuring might emerge from either switching the strategy used to search for moves to attempt (e.g., MacGregor, Ormerod, Chronicle, 2001) or from relaxing self-imposed inappropriate constraints on the problem representation (Knoblich, Ohlsson, Haider, & Rhenius, 1999). Studies of meta-cognition indicate that strategy switching can be unconscious (Newton & Roberts, 2005; Reder & Schunn, 1996; Siegler & Stern, 1998) and that different strategies compete for activation during the strategy selection process (Siegler & Stern, 1998). The conscious- and unconscious-work accounts generate different predictions concerning the effects of activities that individuals engage in during an incubation period. According to the conscious-work account, individuals benefit most from an unfilled incubation period, as this gives them an opportunity either to relax, reduce fatigue, or to continue working on the problem. In contrast, unconscious work accounts suggest that unconscious problem-solving processes occur when individuals shift their attention away from the problem to other mental activities. Thus, a certain level of involvement in other tasks during an incubation period should facilitate post-incubation problem-solving.
A number of experimental studies have examined the role of task type during an incubation period. The experimental paradigms of these incubation studies are fairly uniform: one group of participants is interrupted with an incubation period (having a break or performing other tasks) while solving a problem, whereas the other group works on the problem continuously. Performance differences between these two groups are then compared. The findings of the published studies do not give unconditional support to either the unconscious-work or the conscious-work accounts.
Patrick (1986) found that participants who had a filled incubation period outperformed those who had an unfilled incubation period. However, Browne and Cruse (1988) reported the opposite pattern: participants who took a rest during an incubation period performed better than those who had to perform tasks during an incubation period. There are also studies that report the same level of performance by participants with filled and unfilled incubation periods (Olton & Johnson, 1976, Smith & Blankenship, 1989). However, these studies vary in terms of the length of incubation period, the target problems tackled, and the nature of the interpolated tasks during the incubation period.

However, it is difficult to draw cross-experiment conclusions, since there is no standard operationalization of what constitutes “long” and “short” incubation periods. In Smith and Blankenship’s study (1989), for example, a 15-min incubation period was defined as a long incubation period, and they reported that participants receiving this length of incubation period performed better than those receiving a 5-min incubation period. However, in Beck’s (1979) study, a 20-min incubation period was considered to be short, and participants’ performance in this group did not differ from the control group. Kaplan (1989) suggested that, to judge whether the incubation period is short or long, the length of time that problem solvers spend on initial attempts to solve (named the “preparation period” by Wallas, 1926) should also be taken into account. Kaplan found that a larger incubation effect was observed after increasing the ratio of the length of preparation period to incubation period. Thus, in addition to including incubation and preparation periods as separate moderators in the meta-analysis reported below, a secondary analysis was also undertaken using the ratio of preparation to incubation time as an alternative moderator.

Studies by Palatano & Seifert (1994) and Seifert, et al. (1995) have found evidence of a Zeigarnik effect in insight problem solving (Zeigarnik, 1927, 1938), where individuals remembered the problems on which they got “stuck” better than those solved immediately. Seifert et al. hypothesized that having a better memory for failed problems might help individuals return efficiently to the problem once relevant new information is encountered during an incubation period, thereby maximizing the chance of solving. Evidence concerning this prediction has been obtained in an empirical study carried by Silveira (1971), showing that problem solvers performed better with longer preparation and incubation periods.

Publications that contained studies relevant to a meta-analysis of incubation were collected through a search of the ERIC, PsycInfo, PsycArticles, and MEDLINE databases using the keyword incubat\, intersected with one of fixation, creativ\, divergent\, insight\, or problem. Then, references given in all the obtained articles were systematically searched for additional relevant publications. There is a concern that studies with statistically significant results are more likely to get published than those without significant results, and this may lead to a biased retrieval of studies. To ameliorate this to some extent, similar literature searches were carried out in the ProQuest Digital Dissertations databases and using Google ScholarTM for retrieving PhD dissertations, unpublished papers, and conference papers concerning the incubation effect. In total, 37 relevant publications were identified and obtained...Of the remaining 29 publications, 20 were refereed journal articles, 8 were PhD dissertations, and 1 was a conference paper. The ratio of the refereed to other studies is 2.2:1, which is within the suggested range of between 128:1 and 1:1 for including unpublished studies in an effort to avoid publication bias (Thornton & Lee, 2000). Most publications included multiple experiments, thereby allowing a reasonable sample size of independent studies (n =117) to be achieved.

In some cases, effect sizes had to be calculated from t- and f-values, If a p-less-than value was given instead of a exact p-value, the p-less-than value was treated as an exact value, and an estimate of Cohen’s d was generated. For studies that did not include any of the above-mentioned information but only provided statements of non-significant differences between the control and the incubation groups, then Cohen’s d was assumed to be zero. Among the studies that included multiple incubation conditions, some provided a statement of non-significant performance differences among the incubation conditions, and only reported the overall performance difference between the control and the incubation conditions. In such cases, all incubation conditions were assumed to generate the same magnitude of incubation effect sizes. Of the 117 effect sizes, 88 were extracted directly from the means and standard deviations, t-value, f-value, frequencies, or p-value; 8 were computed from a p less than value; and 21 were estimated from statements of significance.

In some incubation studies, problem solving performance was assessed along more than one dimension. For example, in the study carried out by Vul and Pashler (2007), participants’ performance on RATs was measured in terms of the time spent on solving RATs and the number of correct solutions. In such cases, a single effect sized was computed by averaging the effect size from each measure (cf. Durlak & Lipsey, 1991).
Following Hedge and Olkin’s (1985) suggestion for removing bias caused by small sample studies, an unbiased effect size estimate was computed by multiplying the effect size of each single study by a factor 1-3/(4(total N-2)-1), where total N is the total number of participants of that study.
Any unbiased effect size larger than 2 standard deviations from the group mean was considered an outlier, and was recoded to the value of the effect size found at 2 standard deviations, following a procedure for reducing the bias caused by extreme effect sizes reported by Lipsey & Wilson (2001).

One hundred and seventeen studies were included in this meta-analysis. The total number of participants was 3606, and the median number of participants per study was 25. An unbiased effect size estimate was computed for each independent study. Among these studies, 85 of them report positive effect sizes. The unbiased effect size estimates range from -.71 to 4.07, and the median was .26. The unweighted mean of the unbiased effect size estimate was .41, with a standard deviation of .71.
The upper and the lower bound 95% confidence interval were .54 and .28. Unbiased effect sizes larger than 2 standard deviations from the mean were recoded to the value of the effect size found at 2 standard deviations. Table 3 gives the stem-and-leaf display showing the distribution of the unbiased effect sizes. The unweighted mean of the adjusted unbiased effect size estimate was .36, with standard deviation .51. The upper and the lower bound 95% confidence interval were .26 and .45. The confidence interval does not include zero, implying that the estimate of mean unbiased effect size is significantly larger than zero.
[Insert Table 3 about here]
The heterogeneity statistic, Cochran’s Q, was 173.99, and was significantly larger than the Chi-Square critical value, df = 113, p < .001. This supports the use of random effects model. The variance of each unbiased effect size in random effects model was the sum of the between-studies variance and within-study variance of the unbiased effect size. The between-studies variance, also called the random variance component, among these incubation studies was .0834. The mean of the weighted unbiased effect size was .29, with .04 standard deviation 3 , and the 95% confidence interval was (.21, .39). The non-zero confidence interval implies that the weighted mean is significantly larger than zero. This answers our first question, showing the existence of a positive incubation effect. Figure 1 presents the funnel plot of sample size against estimated unbiased effect size of each study in the meta-analysis.
A weighted least-squares regression using unbiased effect sizes weighted by the inverse of the variance as the dependent variable, and sample size as the predictive variable, was carried out. The regression coefficient of the predictive variable was not significantly different from zero, standardized ! = -.08, p = .41, suggesting the absence of publication bias. Thus no correction has been made for publication bias.
[Insert Figure 1 about here]
Table 4 presents the weighted mean, standard deviation, 95% confidence interval, and random variance component in each sub-group of each categorical moderator.
[Insert Table 4 about here]
Six of the sub-groups (linguistic problems, creative problems, absence of misleading cues, absence of relevant cues, high cognitive load task, unoccupied incubation period) had larger-than-zero random variance components. New weightings, under the assumption of a random effects model, were computed for each of the sub-groups. Weighted least-squares regression analyses were carried out to find the moderators that accounted for the effect size variability among these sub-groups. Small numbers of studies using creative tasks and studies having unoccupied incubation periods preclude the possibility of regression analyses with these moderators. An effect of applying a weighting to this regression analysis is to under-estimate the original standard error of each unstandardized coefficient. Thus, an adjusted standard error was computed by dividing the original standard error by the square root of the mean square residual, a procedure suggested by Lipsey & Wilson (2001). The corrected standard error was used in the significance test (z-test) of each unstandardized coefficient.

With the sub-group of studies using linguistic problems, low cognitive load tasks generated larger incubation effects than rest alone, ! = .54*, p <.05. Also, there was an interaction between Problem Type and Incubation Task with this sub-group, such that that low cognitive load tasks facilitated the incubation effect only when solving linguistic problems.

The negative coefficients associated with “visual problem” and “linguistic problem” indicate that individuals solving these two types of insight problem showed a smaller incubation effect than individuals solving creative problems. A z-test was carried out to compare the coefficients of “visual problem” and “linguistic problem”. The result was not statistically significant, z-score = -1.25, p > .05, suggesting the magnitude of the incubation effect for visual and linguistic insight problems was comparable.
[Insert Table 9 about here]
The length of preparation period was found to have a significant impact on the magnitude of the incubation effect, ! = .03, p < .05. Three bivariate correlations were carried out to check for positive relationships between the length of preparation period and the magnitude of the weighted incubation effect when solving the three types of problem. There was a statistically significant positive correlation between the weighted incubation effect size and the length of preparation period with visual problems, r(35) = .40*, p = .02, and creative problems, r(14) = .60, p = .03, but not with linguistic problems, r(65) = -.04, p = .75.

With visual problems, the magnitude of the incubation effects was independent on the setting of an incubation period (filled or unfilled). Differences between visual and linguistic insight problems may arise through a greater reliance on strategic search rather than knowledge activation in the former than the latter. MacGregor, Ormerod, & Chronicle (2001) proposed that in solving the nine-dot problem, individuals select and execute moves that maximally reduce the distance between current and goal states, essentially drawing lines that connect as many dots as possible. While there remain moves available that satisfy a criterion of satisfactory progress (in this case, the ratio of dots cancelled to lines available), individuals will persevere with an initial representation of the problem that, in the case of the nine-dot problem, does not include consideration of space outside the dot array. According to MacGregor et al’s account, individuals must experience a failure to find moves that meet a criterion of satisfactory progress before they change the initial representation of the problem, thereby including space outside the dot array in their attempts. An incubation period would be helpful only if the problem solvers became aware of the necessity of a strategy shift, but according to MacGregor et al they are unlikely to do so unless they encounter criterion failure as a result of reaching impasse. Siefert, Mayer, Davidson, Palatino, & Yaniv (1995) offer an alternative account that also points to the criticality of experiencing failure and impasse for eventual success in insight problem-solving. If the hypothesis that visual problems require impasse for a strategy switch to occur is correct, a long preparation period (i.e., pre-incubation problem-solving) should be more likely to yield benefits from subsequent incubation with visual problems because it allows individuals to reach impasse prior to incubation.
The results of the regression analysis and the follow-up bivariate correlations are consistent with this prediction, showing a statistically significant positive correlation between the incubation effect size and the length of preparation period with visual problems.
A positive correlation between length of preparation period and incubation effect size was also found with creative problems. A long preparation period may allow individuals to exhaust search in one domain, making it more likely for them to explore a new domain in the second phase of solving. A positive correlation was not found when solving linguistic problems, though this may simply reflect the small variability in length of preparation period among studies using linguistic problems (the preparation period of 82% of these studies were ranged from 0.5 to 1 minute).

While the conscious work hypothesis receives little support here, the meta-analysis leaves open the possibility that unconscious processes may reflect forgetting, activation of new knowledge, or restructuring.
Further experimental studies might focus on comparing the occurrence of these different unconscious processes during an incubation period in different experimental settings. One methodology that might allow such comparisons was employed by Sio and Rudowicz (2007), who examined the occurrence of spreading-activation by measuring individuals’ sensitivity to answers of the unsolved RATs before and after a filled and unfilled incubation periods. They found that the spreading-activation process occurred only in the filled incubation period condition and in a fixated mind, though this study did not measure post-incubation period performance.

One remaining problem is the relatively narrow range of problem types that have been explored. For instance, the majority of studies that explore incubation effects with linguistic problems, which is the majority of studies overall, use the RAT. It is unclear that whether the RAT can be considered an insight problem or a linguistic completion task, suggesting it may not be representative of all linguistic problem-solving tasks. Bowden and Jung-Beeman (2003) have found that participants sometimes claimed that they solved RATs with insight, while sometimes reported that they solved them without insight.
Further studies should aim to explore task-specific experimental settings for maximizing the incubation effect with a wider range of tasks. A further research issue of value might also be to explore individual differences in incubation effects. For example, if the role of incubation is to encourage diffused attention, then individuals who show a propensity towards allocating attention broadly (e.g., as measured via field dependence) may benefit differentially from an incubation period. Also, studies have revealed that strategy switching is related to working memory capacity (Geary, Hoard, Byrd-Craven, & Desoto, 2004). Thus, memory capacity may also interact with incubation effects in solving visual problems." #psychology #creativity