Disclaimer: This is an example of a student written essay.
Click here for sample essays written by our professional writers.

This essay is not an endorsement of any political party or statement. UKEssays.com does not accept payment of any kind for the publishing of political content, it has been published for educational purposes only.

Kuhn Theory of Scientific Revolution

Paper Type: Free Essay Subject: Politics
Wordcount: 5435 words Published: 11th Sep 2017

Reference this


Kuhn’s idea about science is that paradigms will shape inquiry. A paradigm is an open ended format that confirms a shared understanding of reality by utilizing a set of assumptions[1]. Prior to the development of a paradigm, random fact gathering and unstructured research will dominate. Once a paradigm emerges, this will structure future inquiry by establishing set assumptions and theories about the world. A paradigm will reign in a field unchallenged as scientists explore the theories and ideas put forth within the paradigm. Thus scientists will use a paradigm to shape their understanding of the world and what methods they consider relevant to use. Theories that exist in a paradigm are meant to be lenses for understanding the world and will aid in knowledge accumulation and paradigm expansion via puzzle solving[2]. Once key assumptions begin to be challenged, a crisis occurs. Crises are noted to be the cause for changes or adoption of new paradigms. This idea of science argues that gradual evolving of the field will occur as elements are found to misalign in the core assumptions of a paradigm and allow for the development of new ideas. It is worth noting that the Popper ideas about falsification do not greatly impact a paradigm.


Lakatos argues that “a research programe is a sequence of theories within a domain of scientific inquiry.[3]” It is assumed that theories are built atop one another. Thus, newer theories will hold value over the older ones. Lakatos prefers to examine shifts in research programes as being either progressive or regressive rather than describing them as a paradigm shift like Kuhn. Lakatos says that changes which enable more explanation than old theories are a progressive shift while a degenerating research programe refers to a programe that is increasingly finding its theories insufficient to explain a phenomena or issue. Scientists are thus encouraged to abandon degenerative research programes for progressive programes because the degenerating programes are failing to describe as much as a progressive one does. Lakatos argues that scientists will use “heuristic”, or a set of understandings drawn from a constellation of beliefs, to create new theories to replace discarded theories[4].

Get Help With Your Essay

If you need assistance with writing your essay, our professional essay writing service is here to help!

Essay Writing Service

A heuristic can be positive or negative. Negative heuristics refer to how the core of a research programe are assumptions that cannot be altered while positive heuristics allow for modification of some assumptions near the negative heuristics core[5]. This leads to an auxiliary or “protective belt” of ideas around the core assumptions which can be tested and falsified by researchers[6]. While rejection of some elements of this protective belt are ok, if a large portion of the auxiliary belt begins to be rejected, the programe will become degenerative. There are limits to this approach as it can allow for stagnation of ideas and slow the acquisition of new ideas and their adoption into a research programe, problems Kuhn is able to avoid by adoption of more dramatic shifts in the paradigm as opposed to Lakatos and the gradual incremental changes he advocates[7].


Normal science is Kuhn’s idea of scientific inquiry. This concept argues that normal science occurs within a paradigm that engages in theorizing and testing ideas within this paradigm over time[8]. This is a gradual and systemic effort to acquire and test knowledge in a structured manner. This differs from applied science which seeks to answer specific types of questions or problems as opposed to those residing within a paradigm. Normal science can gradually aid in the acquisition of knowledge and can be sued to challenge or replace paradigms if issues are found to reside within the existing paradigm. Without normal science, or knowledge acquisition is haphazard and unfocused. This is often compared to random fact gathering where information is obtained but the expansion of our understanding or knowledge does not occur[9].


We should assume we can divide the social world into systemic and non systemic components because this allows us to effectively measure and detect change. This capacity for categorization (be they nominal, ordinal, interval, or ratio categorizations) allows us to measure and operationalize variables and to track changes over time.


King, Keohane, and Verba (KKV) define inference as “The process of using the facts we know to learn about the facts we do not know.[10]”  For KKV, inference is the ultimate goal of all good science[11]. KKV puts forth two types of inference: Causal and Descriptive. Descriptive inference entails generalizing a single case out to a full population, transfer observable facts and apply them to intangible concepts, and to separate systematic and random components of any phenomena or event[12]. Causal inference  reaches a conclusion about a relationship between X and Y by examining the response of Y when X is changed, indicating that a given change in X will correspond with a given change in Y[13]. This change in X is known as the treatment effect.


The fundamental problem of causal inference is that a counterfactual can never be observed[14]. That is, an observer can never see more than 1 event. Since only one of an infinitely possible number of events can be observed, the researcher or observer can never know the true difference between a control and treatment group in an observational setting as the sample either receives a treatment or does not.


The central limit theorem allows researchers to understand if their sample is statistically significant via an analysis of the distribution of their sample of a population. An increasingly large random sample (larger N) will approach what is known as a normal distribution (or normality) as the sample size increases.  As more sampling occurs, a greater chance of the correct mean will emerge from the sample population[15]. Thus we see a decline in sample variance. The major pay off of this is that we can then identify when other samples are significantly difference from the means of the population as a whole and through the use of z scores estimate the probability of a given sample being the result of random chance.


Statistical significance can tell us the likelihood of a behavior occurring at random. This enables us to reject or accept the causal claims put forth in a hypothesis. This is most useful for causal inference and descriptive inference but will be shaped by our research question and research design (particularly our test).

A type 1 error refers to an error in our measure that gives us a false positive. A false positive is a rejection of the null hypothesis on the grounds that an observation is the result of our treatment when in fact the effect is not actually there. Type 2 error refers to an error in measurement that gives us a false negative which means the results require us to not reject the null hypothesis. This means that our observation indicates that the treatment had no effect when in fact the treatment effect is present. These relate to our error preferences for our inferences in that we must understand what triggered their occurrence. It would be preferable for these errors to be the result of random error than systemic error. This is because a random error would mean these results were just the product of random selection providing us with extreme outliers in this study. If the causes of these errors were systemic, then it would mean that repeated testing would continue to provide us with incorrect results and would obfuscate scientific progress.


Statistical significance tells us the (estimated) probability of a observation occurring by random chance. These measures are useful to us because they allow us to eliminate competing hypotheses from our study. Statistical significance of P at a given value cannot tell us the size of the effect or the direction the effect occurs (though their coefficients can). We should avoid using multiple statistical significance levels when conducting research because arbitrarily applying significance levels of (0.1), (0.05), and (0.01) would allow the researcher to pick and choose what hypotheses they wish to accept or reject. Thus selecting one specific measure (often P=0.05) is preferred.


According to King Keohane and Verba; random measurement does not bias the estimate of causal effect but does make an estimate less efficient or effective[16]. This makes random error in the dependent variables lead to more uncertainty. Random error in the independent variable is noted to also cause uncertainty in the estimates (being too high or too low), but it also leads to a misrepresentation of the causal relationships that may be at play, with some factors being over or under exaggerated in their effectiveness.


According to King Keohane and Verba , systemic error will have no major impact on independent or dependent variables if the measurement error is consistent[17]. For example, if a measure of financial costs consistently adds an erroneous $25 charge, the outcome from the independent variable will not be significantly impact as all units are exposed to the same error. Thus there is no impact on causal inference while descriptive inference will be skewed but still show similar results.


An indeterminate research design is one that yields us no information about the outcome of a hypotheses[18] (KKV118-9). Indeterminate research designs can have more inferences than observations. They can also have two or more explanatory variables that are perfectly correlated with each other[19]. A research design that has 5 units and 12 variables to explore would be a case of an indeterminate research design that would require more units or cases or a reduction in variables (perhaps with a factor analysis). A research design that uses to metrics like income and education could consider replacing one of the metrics to eliminate the multicolinearity.


Selecting on the dependent variable refers to the selection of cases that share a specific feature, phenomena, or outcome that was hypothesized while ignoring those cases where this feature, phenomena, or outcome did not occur. This is problematic as it prevents ones study from being falsifiable, due in part to a lack of variation in the cases selected. This leads to poor inferences and prevents the development of a greater scientific understanding of an issue.


Selection on the independent variable refers to the selection of cases based on the causal mechanisms we wish to study. Selecting on the independent variables does not cause a problem for inference[20]. However, while it does not predetermine the outcome, it will limit the certainty of the conclusions and its generalizability to cases outside the study[21]. This can be limited if we utilized a control variable in the selection process.


In order to craft a case study designed to make strong descriptive inferences, it is best to focus on great depth on the case. We should also limit our cases to as few as possible and utilize process tracing to determine what events happened and to describe them with as much detail as possible and in temporal order. In order to have a case study good for causal inferences, we should make sure that our cases focus on a few features (k) relative to the number of cases (N) in order to make inferences. We must then utilize the four hurdles of causality (making sure there is a relationship between X and Y, making sure y does not cause x, making sure that x and y are covariated, and making sure to eliminate omitted variables) to make sure the measures we use are appropriate in order to make causal inferences from a case study.


Fowler uses an individual’s partisanship relative to the impact of a problem on that person(Fowler, 678) to measure the degree to which an individual will think an election will have an impact on solving that problem[22]. The information needed for this metric was attained by asking the respondents what the most important issue was, which party would do a better job, and if they were affected by the problem. This is an example of a researcher moving from conceptualization toward measurement which is a process in science known as operationaliziation.


Validity refers to the idea that a score utilized by a researcher to measure or score a concept effectively captures the ideas that underlie that concept. There are several types of validity like face (content) validity, which refers to whether a measure captures what it says it does[23]. Construct validity refers to whether a measure is capturing the concept the way it is supposed to[24]. Criterion validity refers to how a measure is related to the tests outcome as the theory would expect[25].

Other forms of validity are internal and external validity. Internal validity refers to the a concept being operationalized in a way where it resembles what it should while avoiding the capture of other variables into itself. Thus a good internally valid measure for altruism will only describe altruism and not other concepts like optimism or religious affiliation. External validity refers to the generalizability to the applicability of the measure (or study) to other issue areas. In the case of this measure of altruism, I would argue that the measure lacks  construct and face validity, as well as internal and external validity. Face validity is failed here as the measure is not effective and thus fails to convey the hypothesized relationship. Construct validity is also not present as the measure is incorrectly measuring altruistic behavior. Criterion validity does hold however as the test makes sense aside from the measure being used being ineffective. Internal and external validity are discussed in 4c.


In my view, the quality of this measure did influence Fowler’s findings. The reason for this is because his measure of altruism appears to capture more of a concern with political affiliation and power rather than a true benevolence or altruistic feeling. Thus the turnout interaction effects that he found to be statistically significant were really just the result of a multiple measuring of partisan behavior. This repeated sampling of partisanship removes internal and construct validity from the operationalization of this concept. Because the measure is not internally invalid (i.e. flawed), it makes no sense for us to try to generalize it outward to other cases because the measure tells us nothing. Thus removing external validity from the operationalization of altruism as well. The combination of partisanship with the dictator game as was done in this experiment contaminates the measure and makes the treatment effect difficult to ascertain as originally intended by the researcher.


In my opinion, I think that Fowler should have stuck with using the dictator game as a measure of altruism in his research. Perhaps he could have pursued a small grant to offset the costs of running the game. Assuming money is not an issue, it would have been worth trying to iterate the game with small variations to more thoroughly sample altruistic behavior. This “battery” of scenarios would mirror Ansolabehere et. al[26] idea about using a battery of related questions to measure a concept, which would be more useful than the measure using a single dictator game coupled with partisan leanings.


According to Kellstedt and Whitten[27] the four hurdles of causality are 1) a credible causal mechanism for x and y, an elimination of the possibility that y will cause x, that covariation exists between x and y, and that confounding variables (z) are eliminated so the x->y connection is not spurious. In this study, I feel that the first and third hurdles are adequately addressed. The problems arise with the second and third hurdles. There is little done to satisfy this readers concerns that perhaps turning out to vote may drive up altruism (due to exposure to others). This needs to be addressed. However, due to poor operationalization, the final hurdle is a definitive problem. The poor operationalization will lead to plenty of possible influence from a z variable (partisanship) that will confound the x-y relationship.


There are several reasons why an observational study may not be a sufficient design to make causal inferences given the specific research question. First, there are many factors that would underlie an individual’s attitude toward a candidate that would not be controllable in an observational setting (such as viewing conditions, volume, focus during commercial breaks). Issues with non random selection, omitted variable bias, and two way causation would be rampant. Though perhaps not a case in this political science subfield, it would be possible that an observational study could be carried out in an area where there are no clear dominant theories (or theories at all) on causal behavior. This would limit any observations to the status of what Kuhn refers to as “random fact gathering”. Additionally, subjectivity by the observers may also prove to be an issue and the lack of control in the setting would just exacerbate selection problems. Lastly, certain types of bias, most notably social desirability bias, will occur if those being observed know that they are being watched. This may skew their response to the negative ads. One could bypass this by observing them covertly but this would introduce numerous ethical problems. It is for these reasons that an observational study may be problematic for addressing the research question of how negative ads influence public support or opposition to a candidate


The standard form hypothesis posits that a change in X will correspond with a change in Y. For this research question a standard form hypothesis would look something like the following: “An increase in the number of negative advertisements about a candidate will lead to a decrease in support of that candidate from a given voter.”


Experiments give researchers leverage by allowing them to control the setting, thus eliminating the omitted variable bias and clearly defining what exposure of the control and experimental groups. This also allows the researcher to precisely determine how the treatment will be applied to the experimental group. This would then allow us to identify if the treatment is responsible for the observed effects without fear of external variables adding error to the results. An experiment also allows the researcher to engage in random assignment rather than hoping for responses from what could be a specific group (such as readers of magazines during an economic downturn vs. the whole population). This would in essence be a controlled experiment instead of a natural experiment. An experimental approach would be more generalizable than observational research. This gives an experiment another advantage as it allows us to say more about the impacts of the causal processes being studied.

Find Out How UKEssays.com Can Help You!

Our academic experts are ready and waiting to assist with any writing project you may have. From simple essay plans, through to full dissertations, you can guarantee we have a service perfectly matched to your needs.

View our services

The experiment would most likely look like a group (N) being divided into 2 halves at random. Group one would serve as the control group and be given a placebo while group two would be the treatment group and receive the treatment. These groups will obviously need to be controlled for gender, education, race, party affiliation, and other demographics to insure each selection is indeed random, and as similar to one another as possible(and representative of the population). The treatment will be negative advertising about a candidate while the placebo will be some other content. The experimenters could then interview or survey the groups after the exposure to ascertain the differences in attitudes and measure the difference to estimate what the treatment effect is.

However if we run our experiment only undergraduate students we run into a few problems with internal and external validity. External validity is threatened because the sample will be so niche and distinct from the rest of the population (younger, more educated, etc) that it will be hard to generalize about the findings beyond this specific sample or the population of young college students. Internal validity also has issues. Such a research question will likely want to specify that the predicted behaviors will apply to likely voters for that specific senate race. By sampling undergraduates we run into a few issues with internal validity. First, there is the assumption that our undergraduates are eligible to vote in this election, which may not be the case if many are out of state students. Additionally there is the issue of lower voter turnout among the youth when compared with older elements of the population. Lastly, there may be lower levels of knowledge (perhaps Converse or Zaller’s ideas about people how select opinions are applicable here  in this tiny niche case) among this sample of the population. These elements all combine to make internal validity unlikely as the sample being measured is unlikely to coincide with the likely voters in the election we are studying.


By targeting only certain channels we may run into a few generalizability problems. Internal validity issues will arise because of how you distributed the ads. Conservative viewers will tend to stick to Fox while liberals will tend to stick with MSNBC. Thus choosing to air ads on only one network may result in other features like partisanship to begin interfering with the later measure of the ads impact. Additionally, without exploring what the local populations look like, we may prevent generalizability (external validity) of our research if the population happens to be skewed in some particular direction. For example, if it is a college town where we run these ads, we won’t be able to generalize to a wider population as those tested would be the youngest of voters while largely omitting analysis of the other age blocks.

We could solve the internal validity problem by controlling for where the ads are visible in a geospatial sense rather than based on channel. This could help avoid problems with viewership patterns that may skew the results, such as pandering to a group that already has a preexisting inclination towards the candidate. If I ran ads on Fox blasting Hillary, it might not have the same effect as if I targeted a general population because the channel is favored by the GOP. Thus the measure would be capturing more of a partisans reaction rather than a general populations behavior. This would also allow for a greater chance of our sample being more normal if we targeted channels in a less specific way. Instead if we chose to air the ads in only specific parts of town while still controlling for other factors in our survey, we could more accurately capture the conceptual impact of negative ads without risking our internal validity. Additionally, participants may hear about the ads without seeing them via conversations with coworkers or other peers, thus the measure may face challenges that we cannot fully control for. Demographics could help generalizability issues however, generalizability is still questionable as this type of experiment cannot be easily repeated, thus limiting the inferences we can draw upon from any study as we wouldn’t be able to tell if this was an outlier case we observed or aligns with a broader trend.

The ethical issues in conducting an experiment like this are a bit trickier. Obviously we would want to obtain IRB approval and ensure we aren’t violating any election laws. However, I am not sure there is an ethical way to do this during an election as our experiment may alter the outcome. This is an ethical issue as our interference is changing something as important as governance. Additionally it introduces a concern about the possibility of the researcher introducing bias or altering the outcomes through interaction with the participants (either through the ads or the surveys).


If we know of in county variation on advertisement rules we can use this to identify the impact of the treatment effect of our ads. Say two cities exist in the county and have differing rules, with one city banning negative ads, and the other allowing them. We can use this to establish a control group and a treatment group and then measure the mean difference in attitudes about the candidate a short time after the campaign. This would allow us to conduct a  natural (real world) experiment with enough control and variation on our causal mechanism to identify whether the treatment had an effect as was posited.


We could use a regression discontinuity design to examine the relationship between victory and funding. Instead of using random assignment we would use a cutoff to eliminate those that score below a margin of victory. We would need to establish a ‘cut off level’ at 50% of voter share as we are in a first past the post system, thus the 50% is a rough estimate for where a majority voting victories will begin. The percentage of the vote attained relative to the amount of funds received would be our scores. The “treatment” in this scenario would be a legislative victory. The treatment effect would be the difference in the amount of money received (if any) between those that won and those that lost around the cutoff point. This would be our localized treatment effect.


Based on the chart we see in figure 2, it appears that we have a sharp RD design. This means that the probability of treatment (in this case receiving enough votes to win the election) is directly tied to the amount of money received has a clear impact around the cutoff and will be consistently P=1[28]. This contrasts with fuzzy RD design where the chance of receiving treatment (electoral victory) would be less than 100% (P<1)[29].

The treatment effect appears to be an increase of around 15 thousand dollars if one is going to win an election. This was determined by measuring the 2 values nearest the cut off and determining the difference by subtracting the ‘control’ group score from the ‘treatment’ group score. This is the treatment effect.

Works Cited

[1] Kuhn, The Structure of Scientific Revolutions, pgs (11-15,23)

[2]Kuhn, The Structure of Scientific Revolutions, pgs (38-39)

[3] https://antimatter.ie/2011/02/11/kuhn-vs-popper-the-philosophy-of-lakatos/

[4] http://www.loyno.edu/~folse/Lakatos.html

[5] http://www.loyno.edu/~folse/Lakatos.html

[6] https://antimatter.ie/2011/02/11/kuhn-vs-popper-the-philosophy-of-lakatos/

[7] http://www.loyno.edu/~folse/Lakatos.html

[8] Kuhn, The Structure of Scientific Revolutions, pgs (10,13,46)

[9] Kuhn, The Structure of Scientific Revolutions, pgs (10,52,64)

[10] King, Keohane, and Verba, Designing Social Inquiry, pg (46)

[11] Brady and Collier, Rethinking Social Inquiry, pg (35)

[12] King, Keohane, and Verba, Designing Social Inquiry, pg (—)

[13] King, Keohane, and Verba, Designing Social Inquiry, pg (—)

[14] King, Keohane, and Verba, Designing Social Inquiry, pg (77)

[15] Kellstedt and Whitten, The Fundamentals of Political Science Research, pg(133-138)

[16] King, Keohane, and Verba, Designing Social Inquiry, pg (156-159)

[17] Ibid.

[18] King, Keohane, and Verba, Designing Social Inquiry, pg (118-119)

[19] Ibid.

[20]King, Keohane, and Verba, Designing Social Inquiry, pg (156-159)

[21] King, Keohane, and Verba, Designing Social Inquiry, pg (137)

[22] Fowler, Altruism and Turnout, pgs (4-5)

Cite This Work

To export a reference to this article please select a referencing stye below:

Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.

Related Services

View all

DMCA / Removal Request

If you are the original writer of this essay and no longer wish to have your work published on UKEssays.com then please: