WEBVTT 00:00:00.900 --> 00:00:03.450 Multiple case study is a very common approach 00:00:03.450 --> 00:00:05.130 to qualitative research. 00:00:05.130 --> 00:00:07.542 In this video, I will talk about Kathleen Eisenhardt's approach 00:00:07.557 --> 00:00:10.625 to the multiple case studies. 00:00:10.650 --> 00:00:12.610 In contrast to quantitative research 00:00:12.610 --> 00:00:14.710 where there are often right and wrong choices 00:00:14.710 --> 00:00:18.250 when it comes to methods, qualitative research is more diverse. 00:00:18.270 --> 00:00:20.820 And it's more difficult to make those choices 00:00:20.820 --> 00:00:23.480 and you really need to understand what you're doing. 00:00:23.480 --> 00:00:29.249 And it requires a lots of expertise to justify your research approach. 00:00:29.260 --> 00:00:34.915 However, to get started people need easy to follow rules and guidelines. 00:00:34.950 --> 00:00:40.082 To this end, many researchers have started to imitate others 00:00:40.100 --> 00:00:41.800 and this has led to the emergence 00:00:41.800 --> 00:00:44.730 of two templates in management research. 00:00:44.730 --> 00:00:48.100 We have the Eisenhardt method and the Gioia method 00:00:48.100 --> 00:00:54.289 and these have been basically emerged, when researchers have started to look at 00:00:54.300 --> 00:00:58.247 who has been able to publish qualitative research in leading journals 00:00:58.250 --> 00:01:01.450 and then started to imitate these authors. 00:01:01.450 --> 00:01:04.650 This Eisenhardt's method is sometimes 00:01:04.650 --> 00:01:08.210 called post positivist, I prefer to call it realist. 00:01:08.210 --> 00:01:11.280 The idea here is that the method 00:01:11.280 --> 00:01:16.280 is more about facts and events that people describe 00:01:16.770 --> 00:01:21.330 instead of being about how people interpret those descriptions. 00:01:21.340 --> 00:01:23.880 So whereas the Gioia method is about understanding 00:01:23.880 --> 00:01:26.710 how people interpret the events around them, 00:01:26.710 --> 00:01:28.120 in Eisenhardt's method, 00:01:28.120 --> 00:01:33.829 we are interested in events themselves and the way people interpret those events 00:01:33.850 --> 00:01:37.190 can produce bias to our analysis results. 00:01:37.190 --> 00:01:42.245 So we try to be more objective whereas Gioia method we try to understand 00:01:42.280 --> 00:01:45.620 the people's subjective use on the events around them. 00:01:45.620 --> 00:01:47.150 There are also other differences. 00:01:47.150 --> 00:01:53.120 In Eisenhardt's method the analysis proceeds a bit differently. 00:01:53.140 --> 00:01:58.380 The approach requires multiple cases, typically four to 10 different cases. 00:01:58.380 --> 00:02:02.240 So for example, if you study power and politics 00:02:02.240 --> 00:02:05.310 which is done in an example paper that I use, 00:02:05.310 --> 00:02:09.810 then you need to pick a couple of organizations where power is, 00:02:09.810 --> 00:02:11.420 where politics is applied 00:02:11.420 --> 00:02:15.110 and couple of organizations where politics is applied less 00:02:15.110 --> 00:02:19.020 and then you analyze those cases and even compare. 00:02:19.020 --> 00:02:21.930 The analysis here in Eisenhardt's method 00:02:21.930 --> 00:02:25.210 is focused on first on within-case analysis. 00:02:25.210 --> 00:02:27.160 So you try to understand each case, 00:02:27.160 --> 00:02:31.909 write the description about the case and then you compare across the cases 00:02:31.950 --> 00:02:36.950 to come up with generalizable theory and propositions. 00:02:37.030 --> 00:02:40.790 So this is much more about objective description and explanation 00:02:40.790 --> 00:02:47.242 of reality instead of being about people's interpretation of that reality. 00:02:47.270 --> 00:02:52.060 These also differ in, how you report the results 00:02:52.060 --> 00:02:55.090 and how you write the paper. 00:02:55.090 --> 00:02:59.658 In Eisenhardt's method there is this quantitative focus. 00:02:59.690 --> 00:03:02.250 Kathleen Eisenhardt's papers quite often 00:03:02.250 --> 00:03:07.657 classify cases according to dimensions as being high and low. 00:03:07.660 --> 00:03:11.699 For example, you can classify the cases into high performing companies 00:03:11.730 --> 00:03:16.574 and low performing companies or companies that are in the middle. 00:03:16.580 --> 00:03:21.198 And then you seek associations between different variables. 00:03:21.200 --> 00:03:23.640 So this is a much more quantitative approach 00:03:23.640 --> 00:03:26.430 to qualitative research than the Gioia method, 00:03:26.430 --> 00:03:30.610 which is more about writing stories on what happened. 00:03:30.610 --> 00:03:34.030 And importantly, the Eisenhardt methods, 00:03:34.030 --> 00:03:38.130 while you develop propositions that are statements 00:03:38.130 --> 00:03:43.947 about causality between two concepts, you need to explain the causal process. 00:03:43.960 --> 00:03:47.520 So she emphasizes that a study, 00:03:47.520 --> 00:03:52.520 a multiple case study must provide explanations, 00:03:52.720 --> 00:03:57.650 not only associations and claims, but also explanations of how, why, 00:03:57.650 --> 00:04:01.190 and when those causal processes work. 00:04:01.190 --> 00:04:02.750 Let's take a look at an example. 00:04:02.750 --> 00:04:05.620 So my example is Eisenhardt and Bourgeios 1988, 00:04:05.620 --> 00:04:08.480 power and politics in high-velocity environments. 00:04:08.480 --> 00:04:13.210 And how does Kathleen Eisenhardt describe their data analysis process? 00:04:13.210 --> 00:04:17.904 Importantly, the data analysis process needs to be described transparently 00:04:17.920 --> 00:04:22.737 because there is no clear guideline or no clear standards 00:04:22.760 --> 00:04:26.010 on how you apply different techniques. 00:04:26.010 --> 00:04:28.540 Whereas in quantitative research you could easily say 00:04:28.540 --> 00:04:31.200 that you apply regression analysis and that's it, 00:04:31.200 --> 00:04:33.540 in qualitative research there's much more freedom 00:04:33.540 --> 00:04:34.820 on how you analyze your data. 00:04:34.820 --> 00:04:38.819 And therefore you need to explain how exactly you did it. 00:04:38.860 --> 00:04:41.840 So the data analysis starts 00:04:41.840 --> 00:04:44.360 by quantifying your data and seeking patterns. 00:04:44.360 --> 00:04:49.430 So quite often in multiple case study following the Eisenhardt approach, 00:04:49.430 --> 00:04:53.950 you have some concepts in your mind before you collect the data. 00:04:53.950 --> 00:04:58.010 And then you simply code for evidence of those concepts 00:04:58.010 --> 00:04:59.680 and different levels of those concepts. 00:04:59.680 --> 00:05:04.000 For example, power and politics in this paper. 00:05:04.000 --> 00:05:07.560 Then you seek associations between power and politics are the correlated. 00:05:07.560 --> 00:05:11.710 If so then you seek evidence for causality. 00:05:11.710 --> 00:05:15.190 Developing profiles is something that Eisenhardt are also recommends. 00:05:15.190 --> 00:05:20.170 So you've write a case description for each case, or it's executive, 00:05:20.170 --> 00:05:22.290 and you can do the same for each decision. 00:05:22.290 --> 00:05:25.380 So you can write profiles, 00:05:25.380 --> 00:05:31.130 short descriptions of the key things or key units that you study. 00:05:31.130 --> 00:05:33.620 Then you develop timelines. 00:05:33.620 --> 00:05:37.730 You can have a piece of paper that has a timeline 00:05:37.730 --> 00:05:40.700 then you put the key events on that timeline. 00:05:40.700 --> 00:05:46.720 And you basically construct causal processes 00:05:46.720 --> 00:05:50.760 or descriptions of some kind of processes using those timelines. 00:05:50.760 --> 00:05:53.080 Then once you have developed 00:05:53.080 --> 00:05:56.070 the case descriptions, developed the timelines, 00:05:56.070 --> 00:05:58.460 then you start to compare pair-wise. 00:05:58.460 --> 00:06:02.520 For example, you could take pairs, you take a high performing company 00:06:02.520 --> 00:06:07.100 a low performing company, you compare and you try to find differences. 00:06:07.100 --> 00:06:11.700 Or if you have two high-performing companies, you try to find similarities. 00:06:11.700 --> 00:06:14.120 And this way you try to find patterns 00:06:14.120 --> 00:06:20.980 that explain, how the cases differ or how they're similar with one another. 00:06:20.980 --> 00:06:26.480 Then you iterate this many, many times and from this a theory emerges. 00:06:26.480 --> 00:06:30.160 So your first seek patterns then based on those patterns 00:06:30.160 --> 00:06:34.882 you start to seek evidence for causality and you iterate many, many times 00:06:34.882 --> 00:06:37.320 because sometimes when you find a pattern 00:06:37.320 --> 00:06:42.310 in for example two cases, that pattern may not exist in other cases. 00:06:42.310 --> 00:06:46.560 And therefore might not be the ideal pattern to follow 00:06:46.560 --> 00:06:49.070 when you construct your theory. 00:06:49.070 --> 00:06:51.450 Finally, you compare with prior research 00:06:51.450 --> 00:06:57.210 to seek for things that are similar and look for things 00:06:57.210 --> 00:07:00.330 that are conflicting with your theory. 00:07:00.330 --> 00:07:04.530 This explanation of data analysis also contains two other important 00:07:04.530 --> 00:07:06.050 and interesting things. 00:07:06.050 --> 00:07:09.230 First, there are conflicts. 00:07:09.230 --> 00:07:12.260 So people will not always tell you the same thing. 00:07:12.260 --> 00:07:17.440 Even if people observe the same events they interpret the events differently 00:07:17.440 --> 00:07:20.960 and people are not objective in how they explain reality. 00:07:20.960 --> 00:07:25.080 Instead they explain their own interpretations of the reality. 00:07:25.080 --> 00:07:26.550 And in Eisenhardt's approach, 00:07:26.550 --> 00:07:29.710 you try to eliminate the influence of a person's interpretation 00:07:29.710 --> 00:07:31.830 of reality from the analysis. 00:07:31.830 --> 00:07:34.370 So if you have conflicting evidence 00:07:34.370 --> 00:07:38.470 then you start looking at what is the reason 00:07:38.470 --> 00:07:39.850 why there's a conflict? 00:07:39.850 --> 00:07:44.550 Why is one person telling us one thing, another person telling us another thing? 00:07:44.550 --> 00:07:47.200 And then based on those conflicting accounts 00:07:47.200 --> 00:07:51.170 you try to infer what is the reality actually like. 00:07:51.170 --> 00:07:55.100 Then another thing that is useful to understand 00:07:55.100 --> 00:07:59.476 is that the patterns are another laws. 00:07:59.476 --> 00:08:05.250 So the fact that most of the cases that applied here, 00:08:05.250 --> 00:08:09.090 that applied politics had centralized power. 00:08:09.090 --> 00:08:13.600 Power centralization does not always lead to the use of politics. 00:08:13.630 --> 00:08:16.940 So you need to understand that it's not a perfect correlation 00:08:16.940 --> 00:08:19.500 like it's never in a quantitative analysis. 00:08:19.500 --> 00:08:23.400 Instead we look for some kind of association 00:08:23.400 --> 00:08:27.800 that may not be always strong they can be weak as well. 00:08:27.800 --> 00:08:29.460 This is how Eisenhardt herself 00:08:29.460 --> 00:08:34.460 describes the process in the seminar, 1989 paper. 00:08:34.700 --> 00:08:37.210 So Eisenhardt's approach starts 00:08:37.210 --> 00:08:40.970 with a research question like we always do. 00:08:40.970 --> 00:08:43.420 In contrast to some grounded theory studies 00:08:43.420 --> 00:08:48.270 where we start from a clean slate, we don't have any ideas 00:08:48.270 --> 00:08:50.830 of what the research result might be. 00:08:50.830 --> 00:08:53.940 In Eisenhardt's approach there is quite often 00:08:53.940 --> 00:08:58.940 some concepts that have been chosen before the study. 00:08:59.180 --> 00:09:02.220 That we want to study the power, we want to study politics. 00:09:02.220 --> 00:09:07.298 So we need to code for evidence of power and evidence or politics in the data. 00:09:08.250 --> 00:09:11.650 Then selecting cases follows theoretical sampling. 00:09:11.650 --> 00:09:15.310 So if we want to study the effects of power and politics 00:09:15.310 --> 00:09:18.370 we should get cases where there's variation 00:09:18.370 --> 00:09:20.580 in power and variation in politics. 00:09:20.580 --> 00:09:24.280 So if possible, we should seek companies 00:09:24.280 --> 00:09:26.850 that we know have centralized power 00:09:26.850 --> 00:09:30.990 and companies that we know where power is not centralized, 00:09:30.990 --> 00:09:34.430 and we can also do this while we sample. 00:09:34.430 --> 00:09:39.930 So let's say that we don't know for sure if cases have centralized power or not, 00:09:39.930 --> 00:09:44.700 and let's assume that we have done four cases 00:09:44.700 --> 00:09:48.690 three have centralized power, in one the power is not centralized. 00:09:48.690 --> 00:09:53.120 Then on the remaining eight cases we should focus on finding companies 00:09:53.120 --> 00:09:55.630 where power is not centralized to the CEO. 00:09:55.630 --> 00:09:59.870 With other constructs at this performance it's much easier because you can get 00:09:59.870 --> 00:10:02.910 objective measures of those concepts 00:10:02.910 --> 00:10:05.990 without actually going in and doing the interviews. 00:10:05.990 --> 00:10:09.380 Then quite often in this approach 00:10:09.380 --> 00:10:12.950 there are multiple ways of collecting data. 00:10:12.950 --> 00:10:17.370 So you collect data through interviews but you can also use survey forms. 00:10:17.370 --> 00:10:23.050 So for example, if you want to study if power is centralized within the CEO, 00:10:23.050 --> 00:10:25.820 you can simply use a power centrality scale 00:10:25.820 --> 00:10:30.108 in the form of a paper-based survey, which you give to the informant 00:10:30.108 --> 00:10:31.740 to fill before the interview. 00:10:31.740 --> 00:10:34.480 So this is a combination of different kinds of data. 00:10:34.480 --> 00:10:39.480 You can use data also from databases like financial data 00:10:40.070 --> 00:10:41.980 and you do the interviews. 00:10:41.980 --> 00:10:47.050 Then importantly data collection and analysis needs to overlap 00:10:47.050 --> 00:10:49.770 always in qualitative analysis because you never know 00:10:49.770 --> 00:10:52.470 what all the concepts are while you start. 00:10:52.470 --> 00:10:55.210 So you might have an idea of the two central concepts 00:10:55.210 --> 00:10:58.814 but if you want to study the process through which power influences politics 00:10:58.814 --> 00:11:01.760 you may not know what the process actually is. 00:11:01.760 --> 00:11:06.300 And once you start to have an understanding, what is the process like? 00:11:06.300 --> 00:11:09.940 Then you can focus your later data collection efforts 00:11:09.940 --> 00:11:11.610 on the important parts of the process 00:11:11.610 --> 00:11:14.310 that you discovered early on in your research. 00:11:14.310 --> 00:11:17.750 Then you do analysis, analysis always starts with within-case analysis. 00:11:17.750 --> 00:11:19.800 So you analyze each case separately 00:11:19.800 --> 00:11:25.000 and then you do cross-case analysis or you do a pair-wise comparisons. 00:11:25.050 --> 00:11:29.510 And for example comparisons using more than two companies. 00:11:29.510 --> 00:11:36.230 Then follows the writing of the tables to show associations and then you write 00:11:36.230 --> 00:11:42.090 the narrative or description of why there are these relationships in the text 00:11:42.090 --> 00:11:49.050 and you enter quotes into the text that explain the process or explain the data, 00:11:49.050 --> 00:11:54.490 show the data that allowed you to infer that there actually is a process. 00:11:54.490 --> 00:11:59.080 And then finally you compare against prior literature 00:11:59.080 --> 00:12:02.780 because this is typically done, 00:12:02.780 --> 00:12:07.780 to come up with a theory between known concepts. 00:12:07.880 --> 00:12:12.410 So it's not very common to apply this kind of multiple case study 00:12:12.410 --> 00:12:16.220 and develop new concepts using multiple case studies. 00:12:16.220 --> 00:12:20.370 So grounded theory is perhaps better for developing new concepts. 00:12:20.370 --> 00:12:25.760 This is more about finding associations and finding causal relationships 00:12:25.760 --> 00:12:30.400 between concepts that you choose before you collect the data. 00:12:30.400 --> 00:12:35.578 And finally, how do you know when do you quit adding more cases 00:12:35.578 --> 00:12:37.395 or stop adding more cases? 00:12:37.588 --> 00:12:40.570 Eisenhardt uses the concept of theoretical saturation. 00:12:40.570 --> 00:12:44.070 So when you have a case that no longer gives you 00:12:44.070 --> 00:12:47.330 more information to guide your theorizing 00:12:47.330 --> 00:12:50.930 then you can conclude that you have reached saturation. 00:12:50.930 --> 00:12:55.030 So collecting more cases would not add much value anymore. 00:12:55.030 --> 00:12:59.140 So you add cases as long as your theory develops 00:12:59.140 --> 00:13:01.950 but once you have the feeling 00:13:01.950 --> 00:13:05.610 that the latest case did not really add much 00:13:05.610 --> 00:13:08.910 then you decide that this is the number of cases 00:13:08.910 --> 00:13:10.770 that I'm gonna use. 00:13:10.770 --> 00:13:14.660 Quite often the number of cases is between eight and 10 00:13:14.660 --> 00:13:16.210 when this approach is followed.